UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON D.C. 20460
OFFICE OF THE ADMINISTRATOR
SCIENCE ADVISORY BOARD
November 14, 2008
EPA-SAB-09-004
The Honorable Stephen L. Johnson
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, N.W.
Washington DC. 20460
Subj ect: SAB Consultation on EPA's Proposed Approach for Estimation of Bin-Specific
Cancer Potency Factors for Inhalation Exposure to Asbestos
Dear Administrator Johnson:
The Environmental Protection Agency's (EPA) current method for quantifying cancer risk from
inhalation exposure to asbestos utilizes exposure measurements based on phase contrast microscopy.
The 1986 method used published epidemiologic studies of miners and manufacturing workers to select
empirical risk models to derive cancer potency factors for lung cancer and mesothelioma. To address
the potential limitations of EPA's 1986 method, EPA's Office of Solid Waste and Emergency Response
(OSWER) has proposed an interim approach to account for the potential differences of cancer potency
between mineral groups and fiber size distributions. The proposed method adopts a "multi-bin"
mathematical model to estimate cancer risk according to mineral groups (amphibole or chrysotile) and
measurements of particle dimensions (length and width) based on transmission electron microscopy
(TEM). OSWER asked the Science Advisory Board (SAB) to conduct a consultation on the proposed
method. OSWER sought SAB advice regarding the soundness of the scientific basis of the proposed
method; the choice of the mathematical models, statistical methods, epidemiologic and exposure data
used; and, what alternative approaches or methods should be considered.
In response to OSWER's request, the SAB Asbestos Committee (Enclosure 1) held a public
meeting on July 21-22, 2008 in Washington D.C. to consider the issue and to provide consultative
advice on the proposed method. An SAB consultation is a mechanism for individual technical experts
to provide comments for the Agency's consideration early in the development of a technical product.
Comments from individual committee members (and subgroups) in response to EPA's charge questions
(Enclosure 2) are included in Enclosure 3. While group consensus was not sought for a consultation, the
Committee would like to underscore major conclusions that emerged from this consultation as described
below.
-------
The general view of the Committee was there is sufficient evidence to support the need for the
Agency's effort in developing risk assessment method(s) to account for potential differences in risk on
the basis of mineral type and size characteristics of asbestos. As detailed in individual and subgroup
comments, there were divergent views regarding whether an effort of this type is warranted at this time.
The Committee, however, generally agreed that the scientific basis as laid out in the technical document
in support of the proposed method is weak and inadequate. A primary concern is the lack of available
data to estimate the TEM specific levels of exposure for the epidemiological studies utilized in this
analysis. The Committee also found that the document was woefully inadequate with respect to the
representation of available information on epidemiology, toxicology, mechanism of action and
susceptibility.
The Committee urged the Agency to support additional targeted research, exposure data
collection and fiber analysis, and validation of alternative risk assessment models. In particular, there is
a critical need for analyses of more epidemiologic studies using TEM based fiber size specific estimates
of exposures as was conducted in the recently published South Carolina textile cohort study. The
ongoing research effort focusing on amphibole asbestos exposure in Libby, Montana would yield
valuable data and insights to further this scientific effort.
The Committee would like to thank the EPA presenters for their expertise, perspectives and
insights that assisted the Committee's understanding of the proposed method. Thank you for the
opportunity to provide early advice on this important topic. The SAB looks forward to receiving your
response and having further interactions as the Agency moves forward in this endeavor.
Sincerely,
/Signed/
Dr. Agnes Kane, Chair
SAB Asbestos Committee
cc: Dr. Deborah Swackhamer, Chair
EPA Science Advisory Board
Enclosures
-------
ENCLOSURE 1
U.S. Environmental Protection Agency
Science Advisory Board
Asbestos Committee
CHAIR
Dr. Agnes Kane, Professor and Chair, Department of Pathology and Laboratory Medicine, Brown
University, Providence, RI
MEMBERS
Dr. Louis Anthony (Tony) Cox, Jr., President, Cox Associates, Denver, CO
Dr. Jeffrey Everitt, Director, Department of Laboratory Animal Science, GlaxoSmithKline
Pharmaceuticals, Research Triangle Park, NC
Dr. Murray Finkelstein, Assistant Professor, Family Medicine Centre, Mt Sinai Hospital, University of
Toronto, Toronto, Ontario, Canada
Dr. Andrew Gelman*, Professor of Statistics and Director, Applied Statistics Center, Department of
Statistics and Department of Political Science, Columbia University, NY, NY
Dr. George Guthrie, National Energy Technology Laboratory, US Department of Energy, Pittsburgh,
PA
Mr. John Harris, Principal, LabCor Portland, Inc, Portland, OR
Dr. Karl Kelsey, Professor, Center for Environmental Health and Technology, Alpert Medical School,
Brown University, Providence, RI
Dr. Paul J. Lioy, Deputy Director and Professor, Environmental and Occupational Health Sciences
Institute, Exposure Sciences Division, UMDNJ - Robert Wood Johnson Medical School, Piscataway, NJ
Dr. Morton Lippmann, Professor, Nelson Institute of Environmental Medicine, New York University
School of Medicine, Tuxedo, NY
Dr. Gary Marsh, Professor, Department of Biostatistics, Graduate School of Public Health, University
of Pittsburgh, Pittsburgh, PA
Dr. Gunter Oberdorster, Professor of Toxicology, Department of Environmental Medicine, School of
Medicine and Dentistry, University of Rochester, Rochester, NY
-------
Dr. Luis Ortiz, Associate Professor of Medicine, Department of Environmental Health, Director
Division of Environmental Medicine, University of Pittsburgh, Pittsburgh, PA
Dr. Julian Peto, Professor, Department of Epidemiology and Population Health , London School of
Hygiene and Tropical Medicine, London,, UK
Dr. Christopher Portier, Associate Director, National Institutes of Environmental Health Sciences
(NIEHS), Director, Office of Risk Assessment Research, NIEHS, Principle Investigator, Environmental
Systems Biology, NIEHS, Research Triangle Park, NC
Dr. Carol Rice, Professor of Environmental Health and Director of the Environmental and
Occupational Hygiene Academic Training Program, Department of Environmental Health, Kettering
Laboratory, University of Cincinnati, Cincinnati, OH
Dr. Randal Southard, Professor of Soils, University of California, Davis, CA
Dr. Leslie Stayner, Director, Epidemiology & Biostatistics, Epidemiology & Biostatistics, School of
Public Health, University of Illinois, Chicago, IL, USA
Dr. David Veblen, Professor of Earth and Planetary Sciences, Department of Earth and Planetary
Sciences, Olin Hall, Johns Hopkins University, Baltimore, MD
Dr. James Webber, Research Scientist, Wadsworth Center, New York State Department of Health,
Albany, NY, USA
SCIENCE ADVISORY BOARD STAFF
Ms. Vivian Turner, Designated Federal Officer, 1200 Pennsylvania Avenue, NW
1400F, Washington, DC
* Dr. Gelman was unable to attend the July 21-22, 2008 public meeting.
-------
ENCLOSURE 2
PROPOSED APPROACH FOR ESTIMATION OF BIN-SPECIFIC CANCER POTENCY
FACTORS FOR INHALATION EXPOSURE TO ASBESTOS
CHARGE QUESTIONS TO THE EPA SCIENCE ADVISORY BOARD
OVERVIEW
At present, EPA uses an approach developed in 1986 for quantifying cancer risk from asbestos exposure
based on phase contrast microscopy as the measure of asbestos exposure. The 1986 method used
existing epidemiological data from cohorts of workers exposed to asbestos in a variety of mining and
manufacturing settings to select quantitative risk models and estimate potency factors for lung cancer
and mesothelioma. EPA's Office of Solid Waste and Emergency Response (OSWER) is proposing an
interim approach to account for the potential differences of cancer potency between different mineral
types and particle size distributions at different human exposure conditions. The document submitted
for review describes a "multi-bin" mathematical approach to estimate cancer risk according to mineral
groups (amphibole or chysotile) and particle size (length and width) based on transmission electron
microscopy. There are a number of issues regarding the statistical methods to be used in the fitting
(these are discussed in Section 8), as well as a number of issues regarding the epidemiological and
exposure data used (these issues are discussed in Sections 9 and 10). The purpose of the following
charge questions is to identify the key issues that OSWER has encountered and to seek input from the
SAB on the proposed approaches for addressing these issues, what changes to the proposed approaches
may be needed, and what alternatives should be considered .
CHARGE QUESTIONS
The proposed approach is based on the hypothesis that there may be significant difference in potency for
lung cancer and/or mesothelioma as a function of asbestos mineral type and particle dimensions.
Charge Question 1:
1. Do you agree that the data are sufficient to indicate that such differences may exist and that an effort
of this type is warranted?
SECTIONS 2-7
Sections 2-5 of the document provide a synopsis on the physical and chemical characteristics of
asbestos, toxicology, epidemiology, and mode of action. An overview of EPA's 1986 dose-response
method is described in section 6, and initial EPA efforts to develop bin-specific cancer potencies are
described in section 7
2-1
-------
Charge Question 2:
2. Please comment on the adequacy of these sections which serve as the scientific bases for the proposed
dose-response assessment approach.
SECTION 8
Section 8 of the document describes the statistical approach that OSWER is proposing for use in fitting
risk models to the available data. Detailed charge questions related to the proposed fitting process are
provided below.
Section 8.2 - Risk Models
OSWER reviewed work done by others in which the adequacy of the risk models for lung cancer and
mesothelioma were assessed. OSWER concluded that the existing risk models (i.e., the same models
developed by USEPA 1986) were adequate for use in this effort.
Charge Questions 3a-3c:
3a. Do you agree that the lung cancer and mesothelioma risk models that are proposed are a
scientifically valid basis for this fitting effort?
3b. Should additional model forms be investigated? If so, what model forms are recommended for
investigation, and what is the basis for concluding that these forms warrant evaluation?
3c. For lung cancer, the current risk model is multiplicative with the risk from smoking and other causes
of lung cancer. Should the nature of the interaction between asbestos and smoking be investigated
further? If so, how should this be done? Do you think the model would be sensitive to additional
quantification of the interaction between smoking and asbestos?
Section 8.3 - Fitting Metric
Fitting of the risk models to the data may occur either at the level of individual studies, or at the level of
individual exposure groups. OSWER is proposing that fitting occur at the level of exposure groups.
Charge Questions 4a-4b:
4a. Is fitting at the group level (based on the number of cancer cases observed) preferred to fitting at the
study level (based on the study-specific KL or KM values)? What are the advantages and disadvantages
ofthis approach?
4b. If so, is it scientifically justifiable to use a Poisson likelihood model for the observed number of
cases in each group? Please comment on any other models that should be considered.
Sections 8.4 - Characterizing Uncertainty In Exposure Data
In most cases, there are multiple sources of uncertainty in the measures of exposure reported in
published epidemiological studies. Section 8.4 provides an overview of how OSWER proposes to
2-2
-------
characterize these uncertainties, and the details of the approach are provided in Appendix C.
Application of the proposed methods to each epidemiological study are presented in Appendix A.
Charge Questions 5a-5d:
5a. Have all of the important sources of uncertainty in cumulative exposure matrices been identified? If
not, what other sources should be accounted for?
5b. Is it appropriate to characterize the uncertainty from each source in terms of an independent
probability density estimated using professional judgment? If not, what alternative approach is
suggested?
5c. Are the general strategies for selecting distributional forms and parameter values described in
Appendix C (and applied in Appendix A) appropriate for characterizing uncertainty in exposure
metrices? If not, what alternative strategies are recommended?
5d. Based on the assumption that each of the sources of error is independent, OSWER is proposing an
approach where the errors combine in a multiplicative fashion. Please comment on the scientific
validity of this approach and provide detailed suggestions for other approaches OSWER should
consider.
Section 8.5. Fitting Approach
OSWER considered a wide range of strategies for fitting the epidemiological data to the risk models,
including simple minimization of squared errors, weighted regression, maximum likelihood methods,
measurement error models, Monte Carlo simulation, and Bayes-MCMC. Based on the recognition that
there is substantial error in both the independent variable (observed number of cases in an exposure
group) and the independent variable (metric of cumulative exposure for the group), OSWER is
proposing Bayes-MCMC as the most robust statistical approach for fitting the data.
Charge Questions 6a-6b:
6a. Is it appropriate to account for measurement error in the exposure data by using "measurement
error" models (weighted regression methods)? If so, how would the weights assigned to each exposure
value be assigned?
6b. Is the assignment of a PDF for data quality sufficient or should data quality be factored into a
weighted likelihood analysis?
6c. Do you think that the proposed strategy of fitting the risk models to the available epidemiological
data using Bayes-MCMC is scientifically justifiable? If not, what alternative strategy do you suggest,
and why?
Section 8.6.2 -Specification of Priors
Assuming that Bayes-MCMC is the method that will be used, it is necessary to specify prior uncertainty
distributions for each of the fitted parameters, including a (the vector of study-specific relative risks of
s
lung cancer at zero exposure), KL (the vector of bin-specific potency factors for lung cancer), and KM
b b
(the vector of bin-specific potency factors for mesothelioma).
2-3
-------
Charge Question 7:
1. Are the priors proposed in Section 8.6.2 for a , KL , and KM consistent with available knowledge?
s b b
If not, what alternative priors should be considered, and why?
Section 8.7 - Comparing Results For Different Binning Strategies
OSWER is proposing an approach in which the best binning strategy is determined empirically (by
finding the strategy that yields the best fit with the data), rather than specifying a binning strategy a
priori that is expected to be optimal based on information from other sources. Conceptually, an infinite
number of binning strategies might be considered. The choice of the size cutoffs for length and width
are judgmental, and are also limited by the availability of particle size distribution data (see Section 10).
OSWER is proposing 20 different binning strategies for evaluation. Length bins proposed for use
include <5, 5-10, and >10 um. Width bins proposed for use are <0.4 and 0.4 to 1.5 um.
Charge Questions 8a-8d:
8a. Do you agree that multiple binning strategies should be evaluated, or do you believe that a
physiological basis exists that can be used to identify a particular set of length and width cutoffs that
should be assessed? If so, what would those length and width cutoffs be, and can these bins be
implemented considering the limitations in the available TEM particles size data sets? (see Section 10)
8b. Are there any of these strategies that you feel do not warrant evaluation? If so, why? Are there any
additional strategies that you recommend for inclusion? If so, why?
8c. Assuming that fitting is performed using Bayes-MCMC, OSWER is proposing that a comparison of
goodness of fit between different binning strategies be based on the Bayes Factor. Do you agree that
this is a statistically valid method for comparing binning strategies? Are there any other comparison
methods you would recommend? If so, why?
8d. Is it important to account for differences in the number of fitting parameters (bin-specific potency
factors) when comparing 1-bin, 2-bin, and 4-bin strategies to each other? If so, how should that be
done?
Section 8.8 - Other Methods For Characterizing Goodness-of-Fit
OSWER is proposing that the initial evaluation of goodness-of-fit of different binning strategies be
based on the Bayes Factor, but is also proposing a number of additional evaluations to assess both
relative and absolute goodness-of-fit. These are described in Section 8.8.
Charge Questions 9a-9e:
9a. What method(s) is (are) preferred for characterizing the absolute goodness-of-fit of any selected
binning strategy? Should any of these methods be used to supplement the relative comparisons based on
the Bayes Factor? If so, how?
9b. If different measures of goodness of fit do not yield results that agree, which method should be
preferred, and why?
2-4
-------
9c. What methodological options do you recommend for validating the results of the modeling efforts?
What are the strengths and limitations of these options compared to others that might be available?
9d. In lung cancer studies, it is expected that the value of a should be relatively close to 1.0. If the fitted
s
value of any particular value of a is substantially higher or lower than 1.0, should this be taken to reflect
s
that the data set giving rise to the value are somehow flawed or are too uncertain for use, and should be
excluded? If so, what criteria would you suggest for recognizing values that warrant concern?
9e. Is an examination performed of the residuals from the meta-analysis a rigorous and scientifically
valid assessment of homogeneity?
Section 8.9 - Sensitivity Analysis
OSWER is proposing an approach for evaluating the sensitivity of the results to the various assumptions
and choices used in the effort that is based on series of "what if tests. For example, this may include
excluding all or some of the data from one or more of the studies, and assessing how those exclusions
impact the results. Likewise, one or more of the PDFs used to characterize uncertain input data may be
changed to evaluate if/how the results are altered.
Charge Questions lOa-lOb:
lOa. Is this "what if approach for evaluating sensitivity scientifically valid and useful?
lOb. Are there other techniques that you recommend for characterizing the sensitivity of the outcome to
the data and methods that are used? If so, what?
SECTION 9. EPIDEMIOLOGICAL DATA PROPOSED FOR USE
Section 9 of the document describes the methods that are proposed for selecting studies for use in the
effort, along with a list of studies that are proposed for inclusion. Detailed charge questions related to
Section 9 are provided below.
Section 9.1 - Criteria For Study Selection
OSWER has reviewed the published literature and identified studies that include sufficient exposure-
response data to allow the study to be included in the model fitting effort for lung cancer and/or
mesothelioma. These rules are as follows:
• The study must be published in a refereed journal.
• The study must provide data that can be expressed in terms of the quantitative risk models for
lung cancer and/or mesothelioma
• The study cohort must consist of individuals who were exposed to approximately the same
atmospheric composition of asbestos.
Some members of the 2003 Peer Consultation panel recommended that a minimum set of data quality
requirements be imposed as part of the study selection procedure, while other members favored
inclusion of all studies and the use of uncertainty factors to account for differences in data quality.
OSWER considered these peer consultation recommendations, and is proposing that no data quality
requirement be imposed because a) formulation of the data quality rules would be very difficult, and b)
2-5
-------
the method for characterizing uncertainty in the data from each study ensures that data from strong
studies has more influence on the results that data from weak studies.
Charge Questions lla-lle:
1 la. Are the study-specific selection rules proposed above scientifically valid for the intended uses?
Should any additional selection rules be added?
1 Ib. Is it appropriate to assume that all workers in a cohort are exposed to the an atmosphere with a
constant composition (i.e., the mixture of asbestos types and sizes is constant) unless the authors report
information to the contrary? If this is not an appropriate assumption, what alternative strategy would be
available?
lie. Should a set of minimal data quality requirements (other than those above) be established for
inclusion of a study in the analysis? If so, what elements of data quality should be considered, and how
should those data quality rules be established?
lid. For lung cancer, OSWER's approach requires that there be at least two exposure groups per study
in order impose some constraint on the value of the study specific value of a. However, OSWER is
proposing to use data from three cohorts described by Henderson and Enterline (1979), even though
there is only one dose group for each cohort. This is because a reliable estimate of a for the combined
cohort can be derived from the data of Enterline et al. (1987). Is this approach appropriate and
scientifically justifiable? If not, can you suggest an alternative strategy for retaining the data from this
important study or should this study be excluded?
lie. One key assumption in any meta-analysis is that the data sets included in the analysis are
homogeneous. How should the assumption of homogeneity be assessed prior to combining the data
from the studies or groups? If you recommend statistical testing, please provide guidance on the
reliability of a decision based solely on the test statistic. If testing produces evidence of heterogeneity
between some studies, what steps can be recommended?
Sections 9.2 and 9.3. Studies Proposed for Use and Studies Excluded
Section 9.2 lists each of the lung cancer and/or mesothelioma studies that OSWER has identified as
being sufficient for inclusion in the data fitting effort. There are a number of studies where cumulative
exposure was not reported in the units needed for modeling. In order to utilize these studies, it was
necessary to use the data provided to estimate cumulative exposure in the needed units (e.g., Yano et al.
2001, McDonald et al. 1982, 1983, 1984). Section 9.3 identifies several studies that were considered for
use, and the reasons why they are proposed for exclusion.
Charge Questions 12a-12c:
12a. Are you aware of any studies that should be included in the model fitting effort that are currently
excluded or omitted? If so, what are these studies, and do they meet the requirements for study
inclusion?
12b. Are there any studies that are currently proposed for inclusion in the analysis that you believe
should be excluded? If so, why?
12c. In cases where the epidemiological data are not reported in the form needed for use in the fitting
effort, are the methods used to estimate the exposures scientifically sound, and are the methods used for
characterizing the uncertainty in the estimates appropriate?
2-6
-------
SECTION 10. METHOD PROPOSED FOR ESTIMATING BIN-SPECIFIC EXPOSURES
One of the largest problems with this effort is that none of the published studies included bin-specific
exposure estimates. Therefore, the effort is contingent upon methods for estimating bin-specific
exposures based on the data provided. Specific charge questions related to this process are provided
below.
Section 10.2 - Extrapolation from Dust to PCM-Based Measures
A number of studies reported exposure in terms of dust rather than asbestos. In some cases, data are
available to extrapolate from dust to asbestos levels. In other cases, no data are provided. OSWER is
proposing to use an "average" extrapolation factor in this case.
Charge Questions 13a-13b:
13a. Is it scientifically justifiable to employ a default dust-to-PCM conversion factor when there are no
site-specific data available?
13b. Are the uncertainty distributions specified in Appendix A to characterize the uncertainty in this
extrapolation consistent with available information and are they statistically appropriate?
Section 10.3 - Extrapolation from PCM to Bin-Specific Measures
The process of extrapolating from PCM-based measures of exposure to bin-specific measures of
exposure requires two types of data: 1) the fraction of the atmosphere that is chrysotile and the fraction
that is amphibole, and 2) particle size data for both the chrysotile and the amphibole components. In the
absence of reliable study-specific data, OSWER is proposing to use published TEM particle size data
from similar workplaces as the basis of the particle size data needed for step 2.
Charge Questions 14a-14i:
14a. Are the point estimates and uncertainty distributions for the fraction amphibole term proposed for
each study scientifically valid?
14b. Is it scientifically valid to use surrogate TEM data to estimate bin-specific concentrations and
exposure values in studies where these data are not reported? If not, what alternative approach could be
followed, or what additional data would be helpful?
14c. Are there any additional bi-variate TEM data sets available that would be useful in this analysis?
14d. Are the point estimates and uncertainty distributions for the fraction amphibole term scientifically
valid?
14e. Can you suggest any ways to improve the process used to identify select the best available
matching TEM data set(s) to a workplace? How sensitive would the model output be to these changes?
14f. Would the model benefit by establishing a common lower cut-point in diameter to normalize the
lower detection limit across studies?
2-7
-------
14g. Do the studies included in the model have surrogate data of sufficient quality and similarity to
expected exposure conditions to support the model? If not, what alternative approach could be
followed?
14h. Are the PDFs described in Appendix C to characterize the uncertainty in the extrapolation of TEM
particle size data from one location to another sufficient and helpful in understanding the implications of
the method used?
14i. Are the extrapolation techniques used on the raw TEM data sets to meet the bin definitions (e.g., 0.4
um diameter) transparent, objectively presented and scientifically valid? Are there alternative techniques
that you would recommend?
SECTION 11 - UTILIZING POTENCY FACTORS TO COMPUTE LIFETIME RISK
Assuming that it is possible to derive a set of bin-specific potency factors, it is expected that these will
be used to evaluate lifetime risk of cancer to an individual with a specified exposure history using the
same basic life-table approach used by EPA (1986). However, each bin-specific potency factor will be
uncertain. Therefore, it is important to specify the uncertainty in the risk predictions that arise from the
uncertainty in the potency factors.
Charge Questions 15a-15b:
15a. What method is best for estimating the uncertainty in lifetime cancer risk predictions that are
associated with the uncertainty in the bin-specific potency factors?
15b. Assuming that estimates of exposure at Superfund sites will also have uncertainty, how should the
overall uncertainty in risk predictions be characterized?
2-8
-------
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. The SAB is structured to provide balanced,
expert assessment of scientific matters related to the problems facing the Agency. 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. Reports of the EPA Science Advisory
Board are posted on the EPA website at http://www.epa.gov/sab.
3-0
-------
ENCLOSURE 3
Comments from Subgroups and Individual Members
Subgroup Comments on Specific Charge Questions
Charge # 1 (Scientific basis for proposed method) — see individual comments
Charge # 2 (Background Information)
Physical/Chemical Characteristics (section 2)—Drs. Veblen, Southard, Gutherie 3-2
Toxicology/Mode of Action (sections 3 & 5) — Drs. Oberdorster, Ortiz, Everitt 3-4
Epidemiology (section 4) — see individual comment (Dr. Rice)
Charge #3 and 4 (Risk Models) — Drs. Lippmann, Stayner 3-14
Charge # 5, 6, 7 (Exposure Data) — see individual comments (Drs. Lioy, Portier, Cox)
Charge # 9, 10 (Statistical Methods and Uncertainty Analysis) — Drs. Portier, Cox, Gelman 3-18
Charge # 11, 12 (Criteria and Selection of Epdemiologic Data) — see individual comments
(Drs. Finkelstein and Stayner)
Charge # 13, 14 (Exposure Measurements and Extrapolation) — see individual comments
(Drs. Harris and Webber)
Charge #15 (Computation of Lifetime Risks) — see individual comments (Drs. Rice and Lioy)
Individual Comments
Dr. Tony Cox 3-20
Dr. Murray Finkelstein 3-24
Dr. Andrew Gelman 3-35
Mr. John Harris 3-37
Dr. Karl Kelsey 3-43
Dr. Paul Lioy 3-46
Dr. Mort Lippmann 3-50
Dr. Gary Marsh 3-53
Dr. Luis Ortiz 3-59
Dr. Julian Peto 3-63
Dr. Christopher Portier 3-66
Dr. Carol Rice 3-71
Dr. Leslie Stayner 3-74
Dr. James Webber 3-82
Additional Materials Supplied by Dr. Peto 3-88
3-1
-------
Subgroup responses to Charge Question 2
Please comment on the adequacy of Sections 2-5 which serve as the scientific bases for the
proposed dose-response assessment approach.
Drs. Veblen, Southard, Gutherie
Section 2 provides background information on asbestos mineralogy, particle size variability, and
measurement methods. These sections should (at a minimum) provide the basis for (1) justification of
the proposed approach and (2) application of the approach (including potential limitations).
Pertaining to justification, the primary points for the proposed binning on mineral type are that:
• Asbestiform amphiboles and chrysotile have different properties that may result in different
biological responses. The current document does not provide sufficient information on this
account.
Information is needed on properties that may relate to biopersistence, which has been shown to
differ between the two groups of minerals. This includes a discussion of solubility and
dissolution rates (e.g., Hume and Rimstidt, 1992, Am. Mineral., 77:1125-1128 and references
therein for chrysotile; Nagy, 1995, Rev. in Mineral., 31:173-234 for chrysotile; Brantley and
Chen, 1995, Rev. in Mineral., 31:119-172 for amphibole) and cleavage/parting that can alter
particle length (e.g., Veblen and Wylie, 1993, Rev. in Mineral., 28:61-137). This discussion
must also include aspects such as dissolution behavior (such as leaching of surface; e.g., Jaurand
et al., 1977, Env. Res., 14:245-254), which impacts the potential release of iron (believed to be
an important factor in some aspects of pathogenesis).
Information is needed on factors related to surface properties, which have been shown to differ
between these materials (e.g., see Veblen and Wylie, 1993). Fubini (1997, Env. Health Persp.,
105:1013-1020) is a good link between surface properties and biological response.
• Asbestiform amphiboles and chrysotile can be differentiated. Section 2.3 addresses this in part.
A helpful addition would be to discuss critically the ability of each technique to differentiate
amphibole from serpentine (chrysotile).
• Binning occupational exposures into the two groups is meaningful. At least two short additions
are needed to allow the evaluation of this: background exposures and nature of mineral deposits.
Klein (1993, Rev. in Mineral., 28:7-59) provides a discussion on factors related to background
exposures, which provides context for assessing studies where the occupational exposures
reported for a given asbestos type were zero.. The nature of mineral deposits is important in
assessing the likelihood of multiple exposures (e.g., some chrysotile deposits contain small
amounts of tremolite asbestos, whereas others do not).
Pertaining to potential application of the approach, it is important to note potential differences between
the exposures used in the assessment and those to which this might be applied. This includes the nature
of the mineral compositions explored and the nature of commercial exposures to amphiboles compared
to exposures to materials from the environment.
• The human data proposed for use relate to a specific set of materials (specifically to chrysotile
and a subset of asbestiform amphiboles). Yet, in the amphibole category, there are mineral
species that are not represented in the proposed epidemiological studies. For example, winchite
3-2
-------
and richterite are two examples of amphiboles that can have asbestiform habits and that are
implicated in asbestos-related disease at Libby, MT (see Wylie and Verkouteren, 2000, Am.
Mineral., 85:1540-1542). Another example is fibrous (not necessarily asbestiform) fluoro-
edenite, which is an amphibole that has been implicated in studies of environmental exposures in
Bianavilla, Italy (Gianfagna et al., 2003, Min. Mag., 67:1221-1229). These amphiboles differ
chemically (and hence in their properties) from those in the proposed epidemiological studies.
Can/should the results of the proposed approach be applied to these other types of environmental
exposures to amphiboles?
• The human data proposed for use relate to materials that are mined and processed. In contrast,
environmental exposures are often to materials that have been exposed to the environment,
which could include factors such as weathering and concurrent exposure to other materials. In
some cases, the environmental conditions may lessen the impact; in others they may augment the
impact. IARC recognizes a distinction in risk between exposures to silica in occupational
settings versus silica exposure under other conditions (in part due to differences in the surface
reactivity). For example, Horwell et al. (2003, Env. Res., 93:202-215) show that generation
potential of free radicals diminishes for volcanic ash following environmental aging. Although
the proposed approach is not assessing any relationships related to occupational versus
environmental exposures, the results may find application beyond the narrow conditions
represented in the epidemiological studies. Hence, a discussion of the differences between the
processed/commercial exposures and environmental exposures would aid in assessing potential
applicability (and/or limitations) of the proposed method.
There are some minor modifications that are also necessary in order to provide a more accurate base:
• Section 2.1-Intro Like it or not, many standard mineralogy texts do not equate "fibrous" and
"asbestiform." To qualify as fibrous, a mineral can merely look fibrous and still be hard, brittle,
and not separable into thin, flexible fragments. It would be better here simply to define the term
asbestiform.
• Section 2.1-Serpentine Modify the sentence on elemental substitutions to note that the most
common substitutions involve small amounts of aluminum and ferrous iron; the additional
elements noted are minor in comparison to these subsitutions. Veblen and Wylie (1993) is a
good reference to serpentine composition.
The word "lattice" in the first paragraph of this section should be replaced by "structure," so that
it reads "in the crystal structure." A lattice is a mathematical construct used to describe crystal
structure but is not the structure itself.
The crystal structure(s) for chrysotile should be described briefly. A description is given for
amphibole, so why not chrysotile?
Section 2.1-Amphibole The first sentence of this section is poor. I suggest replacing it with
"Amphiboles possess double chains of silicate tetrahedra that are interconnected by bands of 6-
to 8-coordinated cations."
The elements listed for the B site should be modified to include Mg and Fe, which are the
dominant B-site elements for the amosite minerals (for example). Veblen and Wylie (1993) is a
3-3
-------
good reference to amphibole? composition. (I would add at least Mn, since this section is
discussing amphiboles in general, not just those that can adopt asbestiform habit.—DRV)
Section 2.1-Amphibole The list of amphibole species should read "grunerite" not "gruenerite".
Also "riebeckite" not "rebeckite."
Though perhaps a matter of taste, some might find it useful to include chemical formulat for these five
amphiboles. (An end-member formula is given in the section on serpentine.)
Last word in section 2.1: The proper mineral name is fluor-edenite, not fluro-edenite.
• Section 2.1 Provide a reference to a more comprehensive discussion of properties, such as the
recent publication by IOM on Asbestos: Selected Cancers.
• Section 2.3-TEM Change "irradiate" to "illuminate" in the first sentence
• Section 2.3-TEM Change "provides the x-ray diffraction pattern" to "provides the electron
diffraction pattern."
• Section 2.3-TEM It is stated that "most TEM instruments" have an accessory that allows the
formation of SAED patterns. In fact, an electron diffraction pattern forms in the back focal plane
of the objective lens whether one likes it or not. It's true that an aperture is required to make this
DP correspond to a specific specimen area, but I know of no TEMs that have ever been
manufactured without at least one SA aperture. Biologists may not use it, but they get it anyway,
whether they like it or not!
Drs. Oberdorster, Ortiz, Everitt
Sections 3-5 are designed as a backbone to illustrate the needs of the OSWER report. These sections
should facilitate analysis of the OSWER document by the members of SAB and the broader public. A
potential consequences of endorsing such document is that the general public may equate this support
with that of an official change in EPA policy toward asbestos carcinogenic potency. It may be advisable
that such document be supplemented to cover the following items.
1. There must be ample discussion in the introduction of these sections stating that this is an
interim approach to assess the question of how physical difference in the composition of asbestos fibers
modify its cancer inducing capacity. A clear description of the historical and current EPA's needs that
motivate the OSWER report should be stated. Specifically, the document should make reference to the
current EPA priorities in the clean up efforts of superfund sites such as Libby, Montana and other sites
around the US (as discussed during the SAB meeting in Washington, D.C.).
2. A more detailed discussion of the importance of the physical aspects of the asbestos fibers
(length as well as width) and whether or not these properties bear directly in the carcinogenic (lung or
mesothelioma) capacity of the asbestos fibers is necessary.
3-4
-------
3. The report should also clearly state that currently there is a definitive paucity of scientific
information, both in animals (Stanton and Wrench , 1972) as well as humans (Stayner LT et a/., 2007)
regarding the use of TEM to characterize fiber size specific asbestos exposure. In the case of Libby
there are samples that could be subjected to TEM analysis to properly address this deficiency.
Therefore, this aspect could be incorporated in the report as a scientific and investigational priority to
the agency.
4. Similarly, there appears to be room for improvement regarding the description of a body of
published work describing the factors that modify the environmental host interaction and determine
individual susceptibility to asbestos-induced cancer. Specifically, there is little description of
epidemiological data addressing the relationship (additive versus multiplicative) between smoking and
asbestos exposure on its carcinogenic effects.
5. Finally, the approach to the biology of the carcinogenic mechanisms of asbestos is timid,
lacking in molecular depth and not providing a biologic foundation to back the epidemiologic approach
map (that could be experimentally adopted to support the imminent fitting of the human epidemiological
data) to indicate how the proposed differences in the physical properties of the asbestos fibers determine
their carcinogenic potency.
The toxicology section is wholly inadequate as described as it doesn't address the important role of
biopersistence in fiber-induced carcinogenesis. It fails to address the synthetic vitreous fiber database
which points strongly to the role of fiber biopersistence as being an important factor in fiber potency and
cancer risk. The fiber biopersistence issue for chrysotile and amphiboles has been reviewed (most
recently by Bernstein and Hoskins, 2006) and should be discussed in this light.
There is also need to review the refractory ceramic fiber studies to discuss the fiber lengths that
were associated with rodent fiber-induced disease as this will emphasize the relatively short nature of
the 10mm proposed bin and the inadequacy of the existing exposure database in the epidemiology
studies.
Another major gap is the lack of discussion of the role of mixed dust exposures on the
pathogenicity of any given fiber exposure. I think the study of Davis and colleagues (1991) should be
specifically referenced with respect to this.
3-5
-------
The charge question gives the impression that the cancer risk assessment approach described in
the document is based on a Dose-Response assessment model. Although that would indeed be the most
desirable approach, it was never attempted anywhere in the document, but rather the analysis is based on
Exposure-Response. There is a significant difference between these 2 approaches, an exposure
concentration is not a dose, and only the dose is obviously most directly correlated with the response.
Figure 1 depicts the interrelationship between Exposure, Dose and Response and some of the important
factors involved. A more detailed description is provided in an earlier review (Oberdorster, 2003,
attached). Deposition, nasal filtering capacity, breathing parameters, fiber length and diameter, and
clearance behavior are among the important factors that determine dose; in particular the deposition and
clearance characteristics of inhaled fibers have been well described by C.P. Yu's group (Dai and Yu,
1998; Yu etal, 1990; 1991; 1998) and most recently by Balashazy et al. (2005). The document suffers
from not discussing these important very basic issues.
It would also be helpful to mention and briefly describe the classical risk assessment paradigm
(NRC, 1983) for further clarification. An attempt to adapt this paradigm together with risk management
consideration, specifically for asbestos, is shown in Figure 2.
The document text (p.3) emphasizes that OSWER focuses only on epidemiological exposure-
response data, and no attempts are made to integrate results from other sources, including animal data,
mode of action, etc. However, the numerous animal toxicological studies that have been performed over
several decades on different types of asbestos can give important information about differences in
biokinetics and biopersistence (retention halftimes, see attached Table 1) of different asbestos types as
well as providing information about dose-response relationships and the importance of fiber dimensions
for translocation to pleural sites, and translocation pathways. These include several long-term inhalation
studies in rats and hamsters with refractory ceramic fibers including asbestos as a positive control by
Hesterberg and colleagues (1993; 1994; 1995; 1997; 1998; 1999) that describe retention kinetics,
effects, and tissue distribution.
In addition, the classic studies by Stanton et al. (1977; 1981), Pott et al. (1974, 1976), Wagner et
al. (1974, 1984) and Davis et al. (1986, 1988, 1991, 1999) related to different types of asbestos,
including amosite, chrysotile, tremolite should be described as valuable background information on fiber
dimension and associated tumorigenic effects and the importance of durability and biopersistence. A
3-6
-------
caveat of these earlier studies relates to the very high exposure concentrations and doses used, which
even with more benign particles would have resulted in lung overload conditions, including significant
inflammation. Also, the impact of co-exposure of asbestos plus a granular dust, whether a benign
particle type (TiCh) or a cytotoxic one (quartz), on cancer induction (increase of both lung cancer and
mesothelioma) and altered retention and biokinetics in rat studies as reported by Davis et al. (1991),
should be included because they highlight the impact of a mixed dust exposure in exposed human
workers (see studies by Finkelstein, 1983; Rey et al., 1994). This is an important concept to be
mentioned and considered in the document especially with regard to asbestos present in superfund sites
potentially resulting in combined exposures.
More recent data point to the enormous differences in the biopersistence between chrysotile and
amphiboles: Chrysotile from Canada, California and Brazil was cleared from the lungs in rat studies
very rapidly, including fibers longer than 20 jim (Bernstein et al, 2005a; 2008; 2004; 2005b).
Retention halftimes for chrysotile fibers >20 jim ranged from less than 1 day to about 11 days,
indicating their low biopersistence. The authors report also less or no pathological responses in rats
after subchronic and chronic inhalation exposure to chrysotile, and suggest that low level exposure to
chrysotile may not be hazardous and the risk much lower than assumed (Bernstein and Hoskins, 2006).
An earlier study in rats on deposition, clearance and translocation of inhaled chrysotile fibers reported a
fast fiber length dependent pulmonary clearance, with retention halftimes (TVa) from below 10 days to
30 days for fibers up to 8 jim long, and about 100 days for fibers >16 jim in length (Coin et al., 1992).
Because retention was measured only for 29 days post-exposure, estimates of the short T!/2 have a
greater level of certainty - and are consistent with significant chrysotile dissolution - than the 100 day
T!/2, which, however, is also consistent with significant dissolution. Although these studies do not
exculpate chrysotile from being labeled as a known human carcinogen, they point to a lower
carcinogenic potency than amphiboles because of the much lower biopersistence of chrysotile. This
should be considered in the risk assessment process.
One suggestion is to compile data of the tox studies in a table form, if available with inclusion of
doses retained in the lung. Animal to human extrapolation models for fiber deposition, retention and
clearance could be mentioned, as developed by C.P. Yu.
Section 5 focuses mainly on the oxidative stress hypothesis of asbestos-induced tumorigenesis,
which is certainly a major mechanism in high dose studies (all animal data) where significant and
3-7
-------
persistent inflammation is induced. While a number of mechanistic studies are cited, a more structured
approach through listing and brief discussion of 5 major mechanistic hypotheses as proposed by Kane
(1996) would be desirable. (Fiber generated radicals damage DNA; fibers interfere with mitosis; fibers
stimulate proliferation of target cells; fibers provoke chronic inflammation with release of ROS and
cytokines; fibers act as co-carcinogens or carriers of chemical carcinogens).
Either here or in the Toxicology section, the importance of fiber biokinetics should be discussed,
which is different between induction of lung cancer and mesothelioma. Once inhaled fibers have
deposited in the different regions of the lung (see earlier comment on fiber size dependent deposition in
tracheobronchial vs. alveolar region), they need to translocate to pleural sites to induce effects (plaques,
fibrosis, mesothelioma). The accepted dogma is that fibers >15/20 jim are the most carcinogenic ones.
While this may be true for lung cancer (alveolar macrophages cannot fully phagocytose long fibers so
they are not cleared), the longer fibers are also least likely to translocate to the pleura. If lymphatic
translocation pathways are involved as suggested by Jones, (1987), lymphatic clearance is limited by
length and diameter, and this is different for pre-nodal and post-nodal lymph (-9-16 jim long; 0.5 jim
thick; Oberdorster et a/., 1988); thus, long fibers are excluded from this pathway. Indeed, animal
studies and human tissue analyses show that by far mostly short fibers (<5-8 jim) were found in pleural
tissue, and few, if any, fibers longer than 10-15 |im (Gelzleichter et al., 1996; 1999; Boutin et al., 1996;
Churg and Vedal, 1994; McDonald et al., 2001; Sebastien et al., 1980; Dodson et al., 1990; Suzuki and
Yuen, 2001, 2002, 2005). Studies reporting fiber dimensions found in autopsy samples of exposed
workers are summarized in Dodson et al., 2003. These authors also suggested to consider other factors
than length for identifying a hazardous fiber, such as surface properties (charge, area, activity), chemical
composition (Fe, other transition models, biopersistence), physiological factors. Summary findings
from all of these studies showed that chrysotile is less prevalent or even absent in human mesothelial
tissues; and these studies indicate also that even short fibers have a carcinogenic potential for inducing
mesothelioma and contributing to lung cancer.
Other studies raise additional issues: Timbrell (1982) found fiber surfaced area to best correlate
with lung pathology (asbestosis) in exposed workers from different mines, introducing the aspect of the
most appropriate dose-metric, which may not necessarily be fiber number. Dogan et al. (2006)
suggested the influence of genetic disposition as susceptibility factor for mesothelioma. He analyzed
the familial history of erionite-induced mesothelioma in several Turkish villages where high of erionite
levels exist in whitewash slurry used as wall cover on their houses.
3-8
-------
Conclusions from the toxicology, fiber tissue burden and mode of action data summarized in the
foregoing paragraphs are that (/') fiber types of asbestos have different potencies (hazard level) with
respect to tumor induction (e.g., low biopersistence for chrysotile); (/'/') the risks for lung tumors and
mesothelioma are different (e.g., translocation of long fibers to pleura); and (in) short asbestos fibers
should not be considered as harmless (are most prevalent in tumor tissue; inflammation as initiating
condition in tox studies). With respect to different types of asbestos, even among amphiboles
differences in potencies seem to exist, e.g., tremolite, erionite as more hazardous types. Exposure-Dose
relationships ought to be clarified in the document. Issues of confounders, mixed exposures and
susceptibility should be discussed. Regarding bin selection for the risk model calculations, tumor type,
asbestos types, and fiber length categories (<5 jim; 5-20 jim; <15 or 20 (im)seem quite appropriate.
References:
Balashazy, I, Moustafa, M., Hofmann, W., Szoke, R., El Huyssein, A., Ahmed, A-R. Simulation of
fiber deposition in bronchial airways. Inhalation Toxicology 17: 717-727, 2005.
Bernstein DM, Chevalier J, Smith P. 2005a. Comparison of Calidria Chrysotile Asbestos to pure
Tremolite: Final results of the inhalation biopersistence and histopathology examination following
short-term exposure. Inhalation Toxicology 17: 427-449.
Bernstein, D. and Hoskins, J.A. The health effects of chrysotile: Current perspective based upon recent
data. Reg Toxicol Pharmacol 45:252-264, 2006.
Bernstein DM, Rogers R, Smith P. 2004. The biopersistence of Brazilian chrysotile asbestos following
inhalation. Inhalation Toxicology 16: 745-761.
Bernstein DM, Rogers R, Smith P. 2005b. The biopersistence of Canadian Chrysotile Asbestos
following Inhalation: Final results through 1 year after cessation of exposure. Inhalation Toxicology
17: 1-14.
Bernstein DM, Rogers R, Smith P, Chevalier J. 2006. The toxicological response of Brazilian chrysotile
asbestos: A multidose subchronic 90-day inhalation toxicology study with 92-day recovery to assess
cellular and pathological response. Inhalation Toxicology 18: 313-332.
Boutin C, Dumortier P, Rey F, Viallat JR, De Vuyst P. 1996. Black spots concentrate oncogenic
asbestos fibers in the parietal pleura. American Journal of Respiratory Critical Care Medicine 153:
444-449.
Churg A, Vedal S. 1994. Fiber burden and patterns of asbestos-related disease in workers with heavy
mixed amosite and chrysotile exposure. American Journal of Respiratory Critical Care Medicine
150: 663-669.
Coin, P.O., Riggli, V.L., Brody, A.R. Deposition, clearance, and translocation of chrysotile asbestos
from peripheral and central regions of the rat lung. Environmental Research 58: 97-116. 1992.
Dai, Y.T. and Yu, C.P. Alveolar depositon of fibers in rodents and humans. J. AerosolMed. 11 (No. 4):
247-258, 1998.
3-9
-------
Davis JMG, Addison J, Bolton RE, Donaldson K, Jones AD, Smith T. The pathogenicity of long versus
short fibre samples of amosite asbestos administered to rats by inhalation and intraperitoneal
injection. Brit. J. Exp. Pathol. 67: 415-430, 1986.
Davis, J.M., Jones, A.D., Miller, E.G. Experimental studies in rats on the effects of asbestos inhalation
coupled with the inhalation of titanium dioxide or quartz. IntJ Exp />a^/70/72(5):501-525, 1991.
Davis JMG, Addison J, Mclntosh C, Miller BG, Niven K. 1991. Variations in the carcinogenicity of
tremolite dust samples of differing morphology. Annals New York Academy of Sciences 643: 473-
490.
Davis JMG, Jones AD, Miller BG. 1991. Experimental studies in rats on the effects of asbestos
inhalation coupled with the inhalation of titanium dioxide or quartz. Int Journal of Experimental
Pathology 72: 501-525.
Davis JMG, Jones AD. 1988. Comparisons of the pathogenicity of long and short fibres of chrysotile
asbestos in rats. Brit. J. Exp. Pathol. 69: pp.717-737.
Dodson RF, Williams MG, Jr., Corn CJ, Brollo A, Bianchi C. 1990. Asbestos content of lung tissue,
lymph nodes, and pleural plaques from former shipyard workers. Am. Rev. Resp. Dis. 142: 843-
847.
Dogan, A.U., Baris, Y.I., Dogan, M., et al. Genetic predisposition to fiber carcinogenesis causes a
mesothelioma epidemic in Turkey. Cancer Res. 66: (10): May 15, 2006.
Finkelstein, M.M. Mortality among long term employees of an Ontario asbestos-cement factory. Br. J.
lnd.Med.4Q: 138-144,1983.
Gelzleichter TR, Bermudez E, Mangum JB, Wong BA, Janszen DB, Moss OR, et al. 1999. Comparison
of pulmonry and pleural responses of rats and hamsters to inhaled refractory ceramic fibers.
Toxicological Sciences 49: 93-101.
Gelzleichter TR, Bermudez E, Mangum JB, Wong BA, Moss OR, Everitt JI. 1996. Pulmonary and
pleural responses in Fischer 344 rats following short-term inhalation of a synthetic vitreous fiber. II.
Pathobiologic responses. Fundamental and Applied Toxicology 30: 39-46.
Hesterberg,T.W., Axten,C., McConnell,E.E., Oberdorster,G., Everitt, J., Miiller,W.C., Chevalier,J.,
Chase,G.R., Thevenaz, P. Chronic inhalation study of fiber glass and amosite asbestos in hamsters:
Twelve month preliminary results. Env. Health Perspectives, Supplement, 1997.
Hesterberg TW, Chase G, Axten C, Miller WC, Musselman RP, Kamstrup O, et al. Biopersistence of
synthetic vitreous fibers and amosite asbestos in the rat lung following inhalation. Toxicology and
Applied Pharmacology 151: 262-275, 1998.
Hesterberg TW, Hart GA, Chevalier J, Miiller WC, Hamilton RD, Bauer J, et al. The importance of
fiber biopersistence and lung dose in determining the chronic inhalation effects of X607, RCF1, and
chrysotile asbestos in rats. Toxicology & Applied Pharmacology 153(Art. No. TO988522): 68-82,
1998
Hesterberg TW, Miller WC, McConnell EE, Chevalier J, Hadley JG, Bernstein DM, et al. Chronic
inhalation toxicity of size-separated glass fibers in Fischer 344 rats. Fundamental and Applied
Toxicology 20: 464-476, 1993.
3-10
-------
Hesterberg TW, Miller WC, Mast R, McConnell EE, Bernstein DM, Anderson R. Relationship between
lung biopersistence and biological effects of man-made vitreous fibers after chronic inhalation in
rats. Environmental Health Perspectives 102 (Suppl. 5): 133-137, 1994.
Jones, J.S.P. Pathology of the Mesothelium, New York, Springer-Verlag, 1987.
Kane, A. Mechanisms of mineral fibre carcinogenesis. IARC Scientific Publications No. 140, Kane,
Boffetta, Saracci and Wikbourn, eds., IARC, Lyon, 1996.
McDonald JC, Armstrong BG, Edwards CW, Gibbs AR, Lloyd HM, Pooley FD, Ross DJ, Rudd RM.
2001. Case referent survey of young adults with mesothelioma: I. Lung fibre analysis. Ann Occup
Hyg.45: 513-518.
Mast RW, McConnell EE, Anderson R, Chevalier J, Kotin P, Bernstein DM, et al. Studies on the
Chronic Toxicity (Inhalation) of Four Types of Refractory Ceramic Fiber in Male Fischer 344 Rats.
Inhalation Toxicology 7: 425-467, 1995.
Miller BG, Jones AD, Searl A, Buchanan D, Cullen RT, Soutar CA, et al. 1999. Influence of
characteristics of inhaled fibres on development of tumours in the rat lung. Annals of Occupational
Hygiene 43(No. 3): 167-179.
Musselman RP, Miiller WC, Eastes W, Hadley JF, Kamstrup O, Thevenaz P, et al. 1994.
Biopersistences of man-made vitreous fibers and crocidolite fibers in rat lungs following short-term
exposures. Environmental Health Perspectives 102 (suppl. 5): 139-143.
NRC (National Research Council). 1983. Risk Assessment in the Federal Government: Managing the
Process. Washington, DC:National Academy Press.
Oberdorster G, Morrow PE, Spurny K. Size dependent lymphatic short term clearance of amosite fibers
in the lung. Ann Occup Hyg 32((Supplement, Inhaled Particles VI),): 149-156, 1988.
Oberdorster, G. Deposition and clearance of inhaled fibers and extrapolation of fiber effects from
rodent studies in humans. J UOEH25 (Supplement 1): 219-235, 2003.
Pott F, Huth F, Friedrichs KH. 1974. Tumorigenic effect of fibrous dusts in experimental animals.
Environ. Health Perspectives 9, pp: 313-315, 1974.
Pott F, Friedrichs K-H, Huth F. Results of animal experiments concerning the carcinogenic effect of
fibrous dusts and their interpretations with regard to the carcinogenesis in humans. Zbl Bakt Hyg
I.Abt.Orig.B. 162: 467-505 1976.
Rey, F., Boutin, C., Dumortier, P., Viallat, J.R., andDe Vuyst, P. Carcinogenic asbestos fibers in the
parietal pleura. Cell Bio. 85: 311-317,1994.
Rogers RA, Antonini JM, Brismar H, lai J, Hesterberg TW, Oldmixon EH, et al. 1999. In situ
microscopic analysis of asbestos and synthetic vitreous fibers retained in hamster lungs
following inhalation. Environmental Health Perspectives 107(No. 5): 367-375.
Sebastien P, Masse R, Bignon J. 1980. Recovery of ingested asbestos fibers from the gastrointestinal
lymph in rats. Environ. Res. 22, pp: 201-216.
Stanton MF, Layard M, Tegeris A, Miller E, May M, Kent E. 1977. Carcinogenicity of fibrous glass:
Pleural response in the rat in relation to fiber dimension. J. Natl. Cancer Inst. 58: 587-603.
Stanton MF, Layard M, Tegeris A, Miller E, May M, Morgan E, et al. 1981. Relation of particle
dimension to carcinogenicity in Amphibole asbestoses and other fibrous minerals. J. Natl. Cancer
Inst. 67: 965-975.
3-11
-------
Stanton, MF and Wrench C. Mechanisms of mesothelioma induction with asbestos and fibrous glass.
Natl. Cancer Inst. 48:797-821,1972.
Stayner, L.T. et al. Occup Environ Med. 2007 Dec 20; [Epub ahead of print]).
Suzuki Y, Yuen SR. 2001. Asbestos tissue burden study on human malignant mesothelioma. IndHealth
39:150-160.
Suzuki Y, Yuen SR. 2002. Asbestos fibers contributing to the induction of human malignant
mesothelioma. AnnNYAcadSci 982:160-176.
Suzuki Y, Yuen SR, Ashley R. 2005. Short, thin asbestos fibers contribute to the development of human
malignant mesothelioma: pathological evidence. International Journal of Hygiene and
Environmental Health 208: 201-210.
Timbrell V. 1982. Depositon and retention of fibres in the human lung. Annals of Occup. Hyg. 26: 347-
369.
Wagner JC, Berry G, Skidmore JW, Timbrell V. The effects of the inhalation of asbestos in rats. Br. J.
Can. 29, pp: 252-269, 1974.
Wagner JC, Griffiths DM, Hill RJ. The effect of fibre size on the in vivo activity of UICC crocidolite.
Br. J. Can. 49: 453-458, 1984.
Yu, C.P. and Asgharian, B. A kinetic model of alveolar clearance of amosite asbestos fibers from the
rat lung at high lung burdens. J. Aerosol Sci. 21: 21-27,1990.
Yu, C.P., Asgharian, B., Pinkerton, K.E. Intrapulmonary deposition and retention modeling of
chrysotile asbestos fibers in rats. J. Aerosol Sci. 22: 757-763,1991.
Yu, C.P., Dai,Y.T., Boymel,P.M., Zoitos, B.K., Oberdorster,G., Utell,MJ. A clearance model of man-
made vitreous fibers (MMVFs) in the rat lung. Inhalation Toxicology 10: 253-274, 1998.
3-12
-------
Figure 1:
Airborne Fibers and Host Interactions
Exposure-
— Sources
/
\
Workplace
Ambient air
— Physico-chemical
properties
(crystalline; amorphous;
organic; composition)
— Concentration
->Dose
— Deposition
Mechanisms
eposil
Clearance/Retention
(Biopersistence)
—Dosemetric
(number; surface; mass)
-> Response
— Inflammation -
— Fibrosis
— Lung tumor
— Mesothelioma-'
s
>*•
I
—Dimension (length, diameter)
— Human activities/ Susceptibility
Risk Assessment and Risk Management Paradigm
For Asbestos-induced Lung Cancer and Mesothelioma
Type- and Size-Bins
Phvsico-chemical vroverties
Airborne size distrib.
cone, by mass, number
surface properties
Public health/social/
economical/political
consequences
Cancer of Lung
and Pleura
Biological Monitoring
(markers of exposure)
Regulations
Expos. Standards
Experimental
Animals
Prevention /Intervention
Measures
Biomed ./Enaineerina
Biokinetics
Biopersistence
Risk
Calculation
Models
Exposure-Dose-
Response Data
(animal~v human)
(high —> low)
Figure 2
3-13
-------
Subgroup responses to Charge Questions 3 & 4 (Drs. Lippmann and Stayner)
3a) Do you agree that the lung cancer and mesothelioma risk models that are proposed are a
scientifically valid basis for this fitting effort?
If "a scientifically valid basis for this fitting effort" means "a basis that allows correct comparisons of
the risks at different sites", then the answer is that we do "not" have sufficient information to know that
the proposed models will provide valid answers. In a broader sense, the EPA 1986 models were
reasonable starting points for this effort in that Crump and Berman [Aeolus 2003], in their 2003 report
for the EPA, tested the EPA 1986 models using raw data from the South Carolina, Quebec and
Wittenoom, Australia study cohorts, and concluded that these models provided reasonably good
predictions for these studies.
On the other hand, we must recognize that these models were based on the best available science in
1986. Even in 1986, they provided rather marginal fits to the available data in the literature, with
individual study KLs and KMs varying by an order of magnitude or more from the models. With the
additional knowledge gained since 1986, these models no longer provide reasonable representations of
the available data. At the same time, the formulations of some of the modeled variables remain valid,
and can be incorporated into improved models.
3b) Should additional model forms be investigated?
It would also be desirable to test the assumption that cumulative exposure (duration x intensity) is an
appropriate metric for the lung cancer model. This should be done if published data exist to allow
separate modeling of duration and intensity. It may only be possible to conduct such an analysis by
obtaining and combining (pooling) the raw data from these studies, which is one of the research
recommendations that has been made.
New models should reflect: 1) the Stayner et al. 2008 analyses of the Charlestown, SC textile worker
cohort, including the evidence that long, thin chrysotile fibers are particularly influential in lung cancer
causation. This provides a human analog to the already well established results from the 2-year rat
inhalation bioassays with chrysotile and amosite of Davis et al. 1986a,b) for the lung cancer model. In
the amosite bioassay, studies, short amosite fibers produced no lung cancers, while long amosite fibers
produced more lung cancers than UICC amosite. In the chrysotile bioassay, short chrysotile fibers
produced a much lower cancer yield than UICC and the even longer chrysotile fibers, but the short
chrysotile was not as free of long fibers as was intended. With regard to the human experience,
At this point, it may be prudent to assume that long-thin chrysotile fibers are more biopersistent within
the lungs than shorter chrysotile fibers and comparable in biopersistence to amphibole fibers, so the
revised lung cancer model can be assumed to apply to long fibers of both fiber types. The results from
the Charlestown cohort may represent a worst-case example, but would be a scientifically defensible
one.
The model for lung cancer implies that age at asbestos exposure is irrelevant, and this should be tested
in the available data. In addition, two-stage clonal expansion (TSCE) or other biologically motivated
risk models that have been developed specifically for lung cancer should be considered to address the
effects of exposure on changing cancer risks over time. Such analyses can only be conducted if a
pooled rather than a metanalysis approach were used.
3-14
-------
For the mesothelioma model, where fibers must translocate from the lung to the pleura or peritoneal
surfaces, it must be recognized that it really only applies to amphibole fibers insofar as it assumes long-
term retention there (over decades) of fibers. While chronic chrysotle fiber exposures have been
associated with mesothelioma, it is now well-established that such fibers are much less potent in terms
of causing mesothelioma than are amphibole fibers (Berman and Crump 2003, ERG 2003, Hodgson and
Darnton 2000, Lippmann 1988) and the revised model must reflect this knowledge.
3c) For lung cancer, the current risk model is multiplicative with the risk from smoking and other
cause of lung cancer. Should the nature of the interaction between asbestos and smoking be
investigated further?
Yes, this relationship should be further evaluated. The observation that there is a multiplicative effect of
smoking and asbestos is largely based on the results from the study of insulation workers [Selikoff et al.
1968, Seidman et al. 1979]. These workers were exposed to relatively thick amosite asbestos fibers, and
were mostly heavy smokers at a time when most lung cancer was squamous cell cancer in the large
airways. It is known that heavy smokers have increased particle deposition in the tracheobronchial tree
(Lippmann and Albert (1969), and abnormal particle clearance from these airways (Albert et al. (1969).
Since then, cigarettes have changed (lower tar content), most lung cancers are now adenocarcinomas,
and most asbestos exposures are to thinner fibers that deposit primarily in smaller lung airways. Also,
we now know that there is a tremendous variation in the evidence regarding whether this relationship is
multiplicative or additive [see Steenland and Thun 1986 and Vainio and Boffetta 1994]. A recent
analyses of this issue suggests that in fact the relationship may be somewhere between additive and
multiplicative [Wraith and Mengersen 2008].
If so, how should this be done?
A more thorough investigation of chrysotile asbestos exposed workers (smokers and non-smokers)
cancer experience may be productive. The sub-multiplicative effect may be due largely to the inclusion
of ex-smokers among the non-smoking group in many studies. Studies with prospective data in which
lifelong non-smokers were identified at recruitment are rare, and ex-smokers are increasingly common.
Careful evaluation of the quality of smoking histories in different cohorts is an important aspect of
further research that is needed. A submultiplicative effect will reduce the predicted risk in smokers.
However, it will greatly increase it among non-smokers, so in an extrapolation from old cohorts with a
lot of smokers to today's population will increase the overall predicted risk. The predicted overall risk
would not be affected by model or measurement misclassification in a population with the same
smoking habits.
Do you think the model would be sensitive to additional quantification of the interaction between
smoking and asbestos?
The dependence of risk on other parameters (fiber type, size, age, duration, etc.) in the model is unlikely
to be strongly affected by assumptions about smoking.
3-15
-------
4a) Is fitting at the group level (based on the number of cancer cases observed [In answering, we
will assume this refers to subgroups based on differing levels of exposure]) preferred to fitting at the
study level (based on the study-specific KL or KM values)?
The committee was divided on this question. Some believed that it was preferable in those cases
where the subgroups are sufficiently large for statistically valid inference. It provides the opportunity to
compare and contrast differing results that might reveal causal characteristics (length, width, fiber type)
associated with the differing fiber characteristics at a specific workplace operation within each study,
that was reported from analyses of some studies (South Caroina, Quebec and Wittenoom). Subgroup
analyses would be particularly valuable in any in-depth review of the completed epidemiology studies
based on experience at Libby, MT.
Other panelists believed that an approach fitting at the individual level was preferable because it more
fully utilized the data available in the studies. This was particularly true for studies where there is
extensive exposure-response information available for modeling such as the South Carolina, Quebec and
Wittenoon studies that were analyzed in the Aeolus report for EPA in 2003.
What are the advantages and disadvantages of this approach?
If grouping can be done on the exposure-specific results, then there are advantages to an initial fitting on
a group level, while preserving the opportunity to combine the results for the study group as a whole.
4b) If so, is it scientifically justifiable to use a Poisson likelihood model for the observed number of
cases in each group?
Yes, it is scientifically justifiable and appropriate to use a Poisson likelihood model for the analysis of
grouped data. If the individual study group results are modeled, one would probably assume a normal
or more likely a log normal distribution applies. In any case, it will be important to test the adequacy of
the distributional assumption made. While the assumption of Poisson variation may be correct, it should
be evaluated by testing for over-dispersion. If there is over-dispersion, an appropriate change should be
made to allow for extra-Poisson variation. In any case, the number of cases will almost always be small,
and a Poisson likelihood model would most likely be appropriate.
References
Aeolus 2003. Report to U.S. Environmental Protection Agency, Washington, DC 20460. May 2003.
Albert, R.E., Lippmann, M., and Briscoe, W. The characteristics of bronchial clearance in humans and
the effects of cigarette smoking. Arch. Environ. Health 1969; 18:738-755.
Davis JMG, Addison J, Bolton RE, Donaldson K, Jones AD, and Smith T. 1986a. The pathogenicity of
long versus short fiber samples of amosite asbestos administered to rats by inhalation and intraperitoneal
injection. Br. J. Exp. Pathol. 67:415-430.
Davis JMG, Addison J, Bolton RE, Donaldson K, and Jones AD 1986b. Inhalation and injection studies
in rats using dust samples from chrysotile asbestosprepared by a wet dispersion process. Br. J. Exp.
Pathol. 67:113-129.
Eastern Research Group, Inc. (ERG). Report on the Peer Consultation Workshop to Discuss a Proposed
Protocol to Assess Asbestos-Related Risk. Final Report. Prepared for the Office of Solid Waste and
Emergency Response. U.S. Environmental Protection Agency, Washington, DC 20460. May 2003.
3-16
-------
Hodgson j and Darnton A. The quantitative risk of mesothelioma and lung cancer in relation to asbestos
exposure. Ann Occup Hyg 44:565-601.
Lippmann, M. and Albert, R.E. The effect of particle size on the regional deposition of inhaled aerosols
in the human respiratory tract. Am. Ind. Hyg. Assoc. J. ; 1969;30:257-275.
Lippmann, M. 1988. Asbestos exposure indices. Environ. Res. 46:86-106.
Seidman, H., Selikoff, I.J., Hammond, B.C. Short-term asbestos work exposure and long-term
observation. Ann. N.Y. Acad. Sci. 1979, 330 :61-89.
Selikoff. I, Hammond E., Churg J. Asbestos exposure, smoking and neoplasia. Journal of the American
Medical Association 1968; 204:104-110.
Steenland. K., Thun M. Interaction between tobacco smoking and occupational exposures in the
causation of lung cancer. Journal of Occupational Medicine 1986; 28:110-118.
Vainio H, Boffetta P. Mechanisms of the combined effect of asbestos and smoking in the etiology of
lung cancer. ScandJ Work Environ Health. 1994 Aug;20(4):235-42.
Wraith D, Mengersen K. Assessing the combined effect of asbestos exposure and smoking on lung
cancer: a Bayesian approach. StatMed. 2007 Feb 28;26(5): 1150-69.
Subgroup Responses to Charge Questions 9 and 10 (Drs. Portier, Cox, Gelman)
9a. What method(s) is(are) preferred for characterizing the absolute goodness-of-fit of any
selected binning strategy? Should any of these methods be used to supplement the relative
comparisons based on the Bayes Factor? If so, how?
The use of Bayes Factors for initial comparisons of different binning strategies seems appropriate, but
additional evaluation methods may be useful. The key question is: When is one binning strategy
unambiguously better than another? Goodness-of-fit tests may not provide clear answers to this
question. Some other techniques that may help are as follows.
(a) You could use conditional independence tests. If binning method A is "more informative than"
method B, in the sense that model predictions are conditionally independent of the information provided
by B, given the information provided by A (but not conversely), then method A is preferable to Method
B.
(b) There is also an opportunity to use simulation-based validation. The Agency could use the
hypothesized relative risk model to generate simulated data for multiple sites, with people having
different distributions of individual exposure histories. Each of the different proposed binning strategies
could be applied to the simulated data. The Agency could then compare the results, and use them to
identify which binning strategies work best (based on criteria such as number of errors or ordinal
correlations between the rank-ordering of the sites using binning and the rank-ordering based on the
detailed simulated data.)
3-17
-------
Finaly, the Agency could do posterior predictive checks, i.e., simulating replicated datasets from the
model and seeing if they look like actual data. To do this, it would be first necessary to create some
useful graphical displays of the data that are being used to fit the model.
9b. If different measures of goodness of fit do not yield results that agree, which method should be
preferred, and why?
This particular question cannot be answered in the abstract since the answer would depend entirely upon
what was seen and why the methods differed. Beyond that, the Agency could use conditional
independence tests and relative performance in comparing simulated sites to provide a different
approach from standard goodness-of-fit that may help to resolve ambiguous cases. Baysian Model
Averaging (BMA) may also prove useful in combining results from multiple plausible models.
9c. What methodological options do you recommend for validating the results of the modeling
efforts? What are the strengths and limitations of these options compared to others that might be
available?
Simulation-based validation discussed under 9a is also applicable here.
9d. In lung cancer studies, it is expected that the value of a should be relatively close to 1.0. If the
fitted value of any particular value of a is substantially higher or lower than 1.0, should this be
s
taken to reflect that the data set giving rise to the value are somehow flawed or are too uncertain
for use, and should be excluded? If so, what criteria would you suggest for recognizing values that
warrant concern?
You are placing too much focus on the nature of this statistic and not asking yourself what it really
means. If this value is substantially different than 1, it means you should return to your data set and
examine why this one data set is so different. Find what that difference is and THEN decide whether it
warrants exclusion from the overall analysis. Does it change your inclusion/exclusion criteria? Do you
need to reevaluate all of the cohorts? The Agency should also focus on ways to explain the differences
that are related to the model structure such as model specification errors, omitted explanatory variables,
and omitted confounders.
9e. Is an examination performed of the residuals from the meta-analysis a rigorous and
scientifically valid assessment of homogeneity?
These types of analyses are not necessarily scientifically (or statistically) valid unless certain
assumptions hold, but they are a reasonable starting point. It would also be useful to also display raw
data and replications of the raw data.
It would also be useful to do these same evaluations after making predictions for new data sets such as
that emerging from Libby, Montana. The Agency could also attempt to apply the model to develop
fiber distributions; these could then be evaluated to see if they are reasonable.
3-18
-------
lOa. Is this "what if approach for evaluating sensitivity scientifically valid and useful?
Sensitivity analyses are indeed a valid and useful technique for understanding the importance of
assumptions on the primary predictions from the modeling exercise. However, many times we approach
these types of exercise without any idea of what we plan to do with the results. The Agency needs to
give some thought to what will be done with the results of the sensitivity analysis.
The Agency has a fairly comprehensive approach to the sensitivity issue. The only additional analysis
we would suggest is that they consider varying the actual form of the model being applied to these data.
Some careful thought should go into this decision for alternative models prior to doing it, since alternate
models may demand alternate prior structures and the Agency would basically be conducting multiple
complete analyses of these data. The purpose of our suggestion is for the Agency to get a solid "feel"
for the impact of alternatives, not necessarily to do multiple full analyses of these data.
Finally, rather than varying one input distribution at a time, try varying the joint distribution of inputs.
For example, one might attempt to solve the optimization problem of choosing joint inputs to the model
(within their allowed or plausible joint distribution) to maximize the prediction error from the model
(using simulated data if necessary.) If the maximized prediction error is small, then this would build
confidence in model predictions.
lOb. Are there other techniques that you recommend for characterizing the sensitivity of the
outcome to the data and methods that are used? If so, what? Model cross-validation provides
another technique with a somewhat similar goal (See Maldonado G, Greenland, (1996) Impact of
model-form selection on the accuracy of rate estimation. Epidemiology 7(1): 46-54).
3-19
-------
Individual Responses to Charge Questions
Dr. Tony Cox
Responses to Charge Questions 1, 5-10, 15
1. Do you agree that the data are sufficient to indicate that such differences may exist and that an
effort of this type is warranted?
This is really more than one question.
To the question "Do you agree that the data are sufficient to indicate that such differences [in potencies
of different fiber types] may exist?", my answer is: Yes.
To the question "Do you agree that the data are sufficient to indicate that an effort of this type [i.e.,
refined risk modeling with exposure variables that take into account differences in fiber characteristics]
is warranted", my answer is: Yes, provided that relevant data on fiber characteristics and risks are used.
However, it may be very desirable to develop additional data to support such refined risk modeling.
To the question "Do you agree that the data are sufficient to warrant the proposed binning strategies as
valid approaches to risk estimation?, my answer would be: Not at present. More data, analysis, and
validation are needed.
A decision analysis perspective may be useful in considering what to do next. It seems very plausible
that different types of fibers (and different types of asbestos) have very different potencies. Failing to
consider these differences could lead to a misallocation of cleanup resources and priorities. The key
decision analysis questions now are:
1. Can EPA allocate resources and set priorities for Superfund cleanups better by using the
proposed binning strategies than by not using them?
2. Would other strategies (e.g., treating fiber characteristics as continuous variables and estimating
the joint distribution of characteristics using nonparametric smoothing or other methods) work
better than binning (and better than ignoring information about fiber characteristics)?
The most useful question is not (or at least should not be) "Does the proposed approach yield correct
answers with high confidence using available data?", but rather: "Does the proposed approach support
more effective risk management and resource allocation decisions in deciding which Superfund sites
should receive highest priority for cleanup?" The answer to the first (irrelevant) question may be no,
while the answer to the second may be yes. Simply ignoring differences between exposures with very
different compositions of asbestos fiber types, posing very different health risks, is almost certainly not
the most effective way to improve EPA's ability to make good risk management decisions based on
available information.
3-20
-------
Many public comments stressed that refined risk analysis is less important than banning activities that
create asbestos risks for workers (or others). In light of these comments, it may be worth emphasizing
the obvious: EPA has asked us about how to use data to set Superfundpriorities most effectively, not
about whether asbestos hazards should exist to begin with.
5 a. Have all of the important sources of uncertainty in cumulative exposure matrices been identified?
(a) No. Other sources include: model form uncertainty, uncertainty about smoking (and other
covariates) that may modify the effects of asbestos exposure, omitted explanatory variables, errors
and biases induced by bin boundaries and discretization, and use of cumulative exposure (which is
probably not a sufficient statistic for predicting risk.)
(b) The relevance of uncertainties about cumulative exposure matrices depends very much on
uncertainties about the true exposure-response relation. Thus, uncertainties about dose-response
(e.g., possible response thresholds or nonlinearities) should interact with uncertainties about
cumulative exposure (e.g., is it above or below a threshold value). (See e.g., Pierce JS, McKinley
MA, Paustenbach DJ, Finley BL. An evaluation of reported no-effect chrysotile asbestos exposures
for lung cancer and mesothelioma. CritRev Toxicol. 2008;38(3): 191-214.) Ideally, uncertainties in
cumulative exposure would be assessed in the context of knowledge and uncertainties in the
exposure-response relation.
(c) There is some uncertainty about whether cumulative exposure provides a sufficient statistic for
predicting risk. For example, suppose that some effects that contribute to induction of lung cancer
or mesothelioma (such as apoptosis or production of specific cytokines in the lung) are
disproportionately triggered by certain concentrations of asbestos (see e.g., Nishimura Y, Nishiike-
Wada T, Wada Y, Miura Y, Otsuki T, Iguchi H., Long-lasting production of TGF-betal by alveolar
macrophages exposed to low doses of asbestos without apoptosis. Int J Immunopathol Pharmacol.
2007 Oct-Dec;20(4):661-71.) Then cumulative exposure metrics may not contain enough
information needed to accurately predict risks. In other words, different detailed exposure histories
that correspond to identical cumulative exposures might have significantly different effects on risk.
(d) Using discrete bins may introduce biases and uncertainties in estimated exposure-response relations
(see e.g., Streiner DL. Breaking up is hard to do: the heartbreak of dichotomizing continuous data.
Can J Psychiatry 2002; 47(3): 262-6.)
(e) Smoking (and perhaps other exposures) may change the effective cumulative exposure metric for
asbestos, e.g., by changing the distribution, retention, and effects of fibers in the lung (Vainio H,
Husgafvel-Pursiainen K, Anttila S, Karjalainen A, Hackman P, Partanen T. Interaction between
smoking and asbestos in human lung adenocarcinoma: role of K-ras mutations. Environ Health
Perspect. 1993 Oct;101 Suppl 3:189-92. Albin M. Poolev FD. Stromberg U. Attewell R. Mitha R.
Johansson L, Welinder H. Retention patterns of asbestos fibres in lung tissue among asbestos cement
workers. Occup Environ Med. 1994 Mar; 51(3):205-11.). Thus, uncertainty about effective
cumulative exposure should incorporate the effects of uncertainty about smoking and other
modulators of asbestos exposure effects.
(f) Variability (interindividual heterogeneity) and uncertainty in cumulative exposure metrics should be
modeled separately.
5b. Is it appropriate to characterize uncertainty from each source as an independent random variable,
using professional judgment? If not, then what instead?
3-21
-------
No. I doubt that uncertainties about these different sources are statistically independent. (For example,
group average exposure durations and group average exposures for unbounded bins might be correlated.
Extrapolations from dust measurements to PCM-based measurements and extrapolations from values
based on PCM to values based on bin-specific concentrations might be correlated with each other and
with use of bin midpoints to represent average exposures.)
Whether professional judgment is valid or useful in this context is an empirical question. It can perhaps
be addressed by independently calibrating different experts and then eliciting joint uncertainty
distributions from them independently, using several different methods (e.g., different factorings of the
joint distribution into products of marginal and conditional distributions). If the different sources are not
considered mutually statistically independent, then copulas or conditional distributions might be used
instead of modeling the uncertainties for different sources as independent random variables. Also,
elicitation techniques can be used that do not assume specific parametric forms for the uncertainty
distributions.
Is there an advantage to quantifying the uncertainty for each source separately, instead of quantifying
uncertainty directly about exposure? If so, can that advantage be captured in the form of known or
assumed algebraic constraints on the possible uncertainty distributions for exposure? This might be
easier and give tighter bounds than quantifying uncertainty separately for each of a bunch of factors and
then combining them. (Here is a simple analogy. Suppose that Y = aX and Z = bY. Each of a and b
may be very uncertain. In that case, quantifying uncertainty directly about the reduced parameter a*b,
e.g., by regressing observed Z values against observed lvalues, may be easier and more informative
than quantifying uncertainty about each of a and b separately and then combining.)
5c and 5d It seems to me that any selection of specific parametric forms for univariate uncertainty
distributions in somewhat ad hoc. Also, expert estimates of different quantities (sources of error) are
not necessarily independent (since an expert who guesses too high on one item may tend to guess high
on others, for example.) Having experts quantify uncertainty distributions directly for the output
(exposure metric) and then for the inputs (perhaps represented as a network of variables from which the
output can be calculated) may give more chances to validate the internal consistency of judgments (and
to resolve any inconsistencies) than simply quantifying univariate distributions for the individual
sources of error and then combining them. (In general, there is no unique way to combine marginal
distributions to get a joint distribution when independence cannot be assumed. Copulas and Bayesian
network representations of dependencies among estimated values of variables may help to better
structure and validate the uncertainty analysis.)
6a-c. Measurement error modeling. I don't think this proposed approach is sufficiently well explained,
as there are several different computational Bayesian methods for missing data that use MCMC and
related methods. I would be interested in seeing a more detailed comparison and evaluation of different
mainstream approaches (e.g., bias-correction algorithms, SEVIEX, EM algorithm, Data Augmentation
algorithm for errors-in-explanatory-variables) and a more detailed rationale for (and explanation of) the
proposed approach.
3-22
-------
References
Berry SM, Carroll RJ, Ruppert D. Bayesian smoothing and regression splines for measurement error
problems. Journal of the American Statistical Association 2002 March; 97(457): 160-169
http://citeseer.ist.psu.edu/309111 .html
Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu C. Measurement Error in Nonlinear Models: A
Modern Perspective, Second Edition Chapman & Hall. New York. 2006
http: //www. stat. tamu. edu/~carroll/ei v. S econdEditi on/index. php
7. Priors. Yes, alternative priors should be used (e.g., uniform over different ranges, exponential, log-
normal, etc.) to reveal the sensitivity of conclusions to the assumed priors. The chosen priors may
be reasonable, but they are surely not the only possible reasonable ones. (A multiple-priors or robust
analysis may also be useful; see http://www.princeton.edu/~noahw/palgrave2.pdf)
8. Binning strategies. Yes, multiple binning strategies should be evaluated. In addition, a binning-
optimization approach might be used (with simulated data, if necessary) to discover what binning
strategies minimize k-fold cross-validation errors. Rather than selecting a single strategy, more
robust results might be achieved by using a Bayesian Model Averaging (BMA) approach that
combines predictions from the several best binning strategies.
9. a and b. Goodness-of-fit and competing binning strategies. Consider using conditional
independence tests (is one binning approach more informative than another?), model cross-
validation and BMA (or bagging) to combine results of multiple binning models, rather than just
using goodness-of-fit criteria and selecting one binning strategy. (When model uncertainty is
considered, as it ought to be, it often turns out that selecting any single model or strategy increases
the risk of erroneous conclusions compared to that from combining results from several plausible
models.)
9c. Validate the modeling approach and results using simulation-validation (i.e., simulate a data set on
fiber characteristics and risks for several sites, apply the selected binning strategy or strategies to the
simulated data, then evaluate and compare the performance of the selected strategies in setting cleanup
priorities among the simulated sites. Note that the correct answers are known for simulated data, since
all details on fiber characteristics and risks are known.)
9d. If alpha is unexpectedly far from 1, consider model misspecification as a possible explanation. (It
may be more sensible to blame the model - and improve it - rather than blaming the data.)
10. What-if approach, (a) Rather than varying one input distribution at a time, try varying the joint
distribution of inputs. For example, one might attempt to solve the optimization problem of
choosing joint inputs to the model (within their allowed or plausible joint distribution) to maximize
the prediction error from the model (using simulated data if necessary.) If the maximized prediction
error is small, then this would build confidence in model predictions, (b) Model cross-validation
provides another technique with a somewhat similar goal.
3-23
-------
15a. How to estimate the uncertainty in lifetime cancer risk predictions from uncertain bin-specific
potency factors?
15b. How to incorporate exposure uncertainties at Super fund sites? One simple framework for
answering both questions is: Use Monte Carlo uncertainty analysis, i.e., sample from the joint
distribution of uncertain model-input quantities (including bin-specific potency factors and exposure
uncertainties), then push the sampled values through the rest of the model (using a tool such as
Analytica) to obtain a corresponding distribution for predicted cancer risks.
Model uncertainty is a more difficult (and probably more important) issue. The basic model of lung
cancer used here is a relative risk model, not a biologically-based model that predicts effects of
exposures on age-specific hazard functions by quantifying effects on underlying mechanistic (e.g., cell
transition and proliferation) rate parameters. To make defensible predictions of effects on lifetime risks
(and uncertainties about those effects), it might be very desirable to use a more biologically-based model
of lifetime cancer risks. (The current report's claim that model specification errors are unlikely to be
significant does not seem to me to be very self-evident. Considering other model forms that more
directly reflect relevant lung cancer and mesothelioma biology might be valuable.) Without considering
how risks change over time, given exposure histories, there may be no sound basis for deciding which
sites to clean first. The goal of cleaning sites is to reduce health risks, so the relation between reduced
exposure and reduced age-specific health risks is crucial for informing EPA's decisions. The relative
risk model is not necessarily the best approach for quantifying this risk. TSCE and other models might
give better results, and should be considered as well.
Dr. Murray Finkelstein
Charge Question 1: Do you agree that the data are sufficient to indicate that such
differences may exist and that an effort of this type is warranted.
I believe that animal and human data suggest that there are potency differences related to
asbestos mineral type and dimension. I also believe that an effort by the EPA to study these potency
differences would be worthwhile.
I do not believe that an effort of the type outlined in the Proposed Approach is warranted because it is
based solely on human epidemiologic data. With the exception of the recently published update of the
South Carolina textile plant, none of the human studies provide the data required to analyse the
proposed model. In essence, all of the input data would consist of "guesses" and the output from the
model would not be credible.
Charge Question 2: Comment on adequacy of these sections which serve as the scientific
basis for the proposed dose-response approach.
Section 4:
This section is an overview of human studies and covers non-cancer and malignant effects. The section
is brief and superficial, but covers outcomes of asbestos exposure in humans.
3-24
-------
The material is generally adequate. I will suggest a few editorial changes:
Asbestosis — Paragraph 2.
The authors write that difficulty breathing is often accompanied by "coughing and rales".
Coughing is a symptom, rales are a sound audible through a stethoscope. Rales are often an early sign
of asbestosis. I would suggest deleting mention of rales (which few non-physicians will understand).
Pleural abnormalities
"Pleural effusions are areas".... no.
Pleural effusions are the fluid which collects in the pleural spaces.
Lung Cancer
"The risk of developing lung cancer from asbestos exposure is substantially higher in smokers than non-
smokers" ... no.
While it is certainly true that smokers exposed to asbestos have a higher overall risk of lung cancer than
do non-smokers, under the multiplicative model, the asbestos-attributable risk of lung cancer is
unrelated to smoking status.
Charge Question 11
1 la. Are the study specific selection rules proposed above (Section 9.1) scientifically valid for the
intended uses? Should any additional selection rules be added?
OSWER must decide whether to exclude internally-controlled studies. The OSWER author has
misinterpreted ALL internally controlled studies, confusing Odds Ratios and Relative Risks with SMRs.
There is no confidence interval for the Reference Category and no value of a for any of these studies.
Criterion 3 is unclear in that "atmospheric composition" is not defined. If atmospheric
composition means the mix of asbestos fibre types, then this is reasonably assessed. If
atmospheric composition means, in addition, the distribution of fibre sizes (length and diameter) then
most study cohorts don't provide enough information to enforce this criterion. In practice, OSWER
proposes to "average over" the varying processes that create the varying fibre-size distributions.
From an epidemiologic perspective, I would propose the addition of the following new rules:
New rules: 1) Source of outcome (mortality) data must be the same for study subjects and the reference
population (not satisfied in Lacquet study)
2) Must stratify or account for latency. Since lung cancer or mesothelioma excesses
are not usually seen before 15-20 years from start of exposure, inclusion of early PYR will dilute
apparent risk.
3-25
-------
1 Ib. Is it appropriate to assume that all workers in a cohort are exposed to the same
atmosphere with a constant composition (i.e. the mixture of asbestos types and sizes is
constant) unless the authors report information to the contrary? If this is not an
appropriate assumption, what alternative strategy would be available?
It is not appropriate to assume that all workers in a cohort are exposed to the same atmosphere with a
constant composition. OSWER acknowledges that the fibre size distribution is process dependent, and
this has been re-demonstrated in the recent work of Stayner and colleagues.
Averaging over work processes, as OSWER proposes to do, will mask the very risk differences that
OSWER is wanting to determine. The best strategy, as implemented by Stayner and colleagues, is to
undertake TEM-based exposure-response analyses. Failing this, I think that the best strategy is ignore
the fibre size question and to compute sector-dependent risks, eg, mining, textile, friction materials etc.
lie. Should a set of minimal data quality requirements (other than those above) be
established for inclusion of a study in the analysis? If so, what elements of data quality
should be considered, and how should those data quality rules be established?
No, I agree that these will be difficult to establish, and that the OSWER approach is acceptable.
lid. For lung cancer, OSWER's approach requires that there be at least 2 exposure groups per study in
order to impose some constraint on the value of the study specific value of alpha. However, OSWER is
proposing to use data from 3 cohorts described by Henderson and Enter line (1979) even though there is
only one dose group for each cohort. This is because a reliable estimate of alpha for the combined
cohort can be derived from the data of Enter line et al. Is this approach appropriate and scientifically
justifiable? If not, can you suggest an alternative strategy for retaining the data from this important
study or should this study be excluded?
I am not sure that there is a reliable estimate for a for the combined cohort.
The Enterline cohort is composed of retirees from a large company with many work locations.
There is no rationale to believe that a single value of alpha is appropriate for all subcohorts.
Also, the deaths are coded to ICD 7.
OSWER has failed to subtract the mesothelioma deaths to get the number of lung cancer cases.
Using linear regression, I calculate a different value for the intercept (116), same as Enterline's.
From Poisson regression log(Obs/Exp) = 0.523 at CumExp = 0. ==> intercept is 1.68. Neither of these is
the same as the alpha displayed on Figure A3-4.
I suspect that the data from chrysotile only and asbestos-cement pipe subcohorts is captured in the
Hughes and Weill study.
3-26
-------
As for values of a in other studies:
1. Berry and Newhouse (Al). This is based upon case-control data. There is no value of alpha.
2. Albin at al (A12): The OSWER author treats this study as a standard SMR-type study, which it is not.
There was a defined industrial reference population of 1233 men, not a national reference population.
Relative risk was computed by age and calendar period adjusted Poisson regression modeling. The data
presented in Figure A12-1 are thus not meaningful and the value of a is not compatible.
3. Libby vermiculite. The OSWER author treats this study as a standard SMR-type study, which it is
not. Comparisons were internal to the cohort, not external to a national reference population.
Relative risk was computed by adjusted Poisson regression modeling. The data presented in Figure
A12-1 are thus not meaningful. There is no measure of "expected deaths" for the referents. There is no
95% confidence limit for the Reference Category and no value of a.
4. Witennoom crocidolite miners. OSWER fails to recognize that this is a case-control study. Figure
A14-1 is thus not meaningful. There is no confidence interval around the Odds Ratio for the reference
category. The concept of observed and expected is not well defined. The 90% confidence Intervals are
probably inaccurate and there is no value of a.
5. China Asbestos Products Factory. The OSWER author treats this study as a standard SMR-type study,
which it is not. Comparisons were internal to the cohort, not external to a national reference population.
Relative risk was computed by Cox proportional hazards regression modeling. The data presented in
Figure A17-1 are thus not meaningful. There is no measure of "expected deaths" for the referents. There
is no 95% confidence limit for the Reference Category and no value of a.
In summary: Many of the values OSWER derives for a are invalid.
lie. One key assumption in any meta-analysis is that the data sets included in the analysis are
homogeneous. How should the assumption of homogeneity be assessed prior to combining the data from
the studies or groups? If you recommend statistical testing, please provide guidance on the reliability of
a decision based solely on the test statistic. If testing produces evidence of heterogeneity between some
studies, what steps can be recommended?
I think that there is a more serious problem. I don't think that the individual data sets are
internally homogeneous. There are different process-dependent fibre size distributions, and OSWER
proposes to smear over them by combining the size distributions from the various processes.
There is also no reason to believe that the inter-study dose estimates are consistent.
Charge Questions 12:
12a: Are you aware of any studies that should be included in the model fitting effort that
are currently excluded or omitted? If so, what are these studies, and do they meet the
requirements for study inclusion?
3-27
-------
New report on South Carolina textile factory by Stayner et al in OEM
(doi: 10.1136/oem.2007.035584) needs to be considered.
Finkelstein has also published a mesothelioma dose-response study:
Finkelstein M. Analysis of the exposure-response relationship for mesothelioma among
asbestos-cement factory workers. Ann NY Acad Sci 1991;643:85-89.
Finkelstein M. The exposure-response relationship for mesothelioma among asbestos-
cement factory workers. Toxicology and Industrial Health 1990;6:623-627.
Italian Balangero Mesothelioma
(Mirabelli D, Calisti R, Barone Adesi F, Fornero E, Merletti F, Magnani C.
Excess of Mesotheliomas after Exposure to Chrysotile in Balangero, Italy.
Occup Environ Med. 2008 Jun 4. )
12b. Are there any studies that are currently proposed for inclusion in the analysis that you believe
should be excluded? If so, why?
The Belgian (Lacquet) study should be omitted because the mortality data for the cases (family doctor or
social workers) are derived from a different source than the reference data (death certificates).
12c. In cases where the epidemiological data are not reported in the form needed for use in the fitting
effort, are the methods used to estimate the exposures scientifically sound, and are the methods used for
characterizing the uncertainty in the estimates appropriate?
See study specific comments below.
Al. British Friction Products Factory (Berry and Newhouse 1983)
Text of Draft Report
Fraction Amphibole: OSWER proposes a screening level value of 0.5% for the average fraction
amphibole in the workplace. However, exposure was likely to be binary (Yes/No). OSWER proposal
thus introduces misclassification.
Estimating Cumulative Exposure: OSWER states that occupational histories were extracted from
employee personnel files and used to estimate levels of cumulative exposure for each individual. This is
not correct. Cumulative exposures were estimated only for subjects in the case-control study.
Smoking Data: OSWER states that the workers demonstrated a reduction in smoking compared to the
national population. This is not correct. Berry and Newhouse speculated in their discussion that a
reduction in smoking might account for lower SMR, but would not influence case-control study.
3-28
-------
Lung Cancer Results: Berry and Newhouse fit a linear model to the case-control data. There was no
significant dose-response relationship.
It must be realized that the dose-response results for this study are based upon case-control and not
cohort data. The Expected deaths in Figure Al-1 are meaningless in the context of a case-control study.
There is no explanation how the author derived his Confidence Limits in the Table. They are
meaningless. There is no confidence interval about the baseline in a case-control study.
Uncertainty in Particle Size data: It is likely that the particle size distribution depends upon the job and
the expenditure of mechanical energy to disrupt the asbestos fibres. Combining data from mixing,
forming, and finishing, averages out all distinctions and is noninformative.
Discussion: This study provides some estimates of risk in a primarily chrysotile environment. There is
little useful information about fibre size distribution. OSWER has misinterpreted this case-control study.
A2. South Carolina Textile Plant (Hein et al, 2007)
Text of Draft Report
PCM - f/cc Conversion factors
Dement and McDonald disagree on conversion factors (by about a factor of 2). These are not reconciled.
Uncertainty in Particle Size Data
Data sets for different operations combined. Recent analysis by Dement et al (OEM, 2007) demonstrate
differences between operational areas. Combining different operations smears out size distribution
differences.
Discussion
This is one of the important studies in asbestos epidemiology. Risk, in relation to fibre dimension data,
is available in updated analyses from Stayner and colleagues.
New report in OEM (doi: 10.1136/oem.2007.035584) needs to be considered.
A3: Retirees from US Asbestos Products Factory
Draft Report Text
Conversion from mppcf to f-yr: Use default factor of 3 for all processes and all jobs
Lung Cancer Results:
Henderson and Enterline coded to ICD 7:
ICD 162 and 163 include mesothelioma (table 3, 1987)
ICD 7 coding table:
3-29
-------
162 Malignant neoplasm of bronchus and trachea, and of lung specified as primary
162.0 Trachea
162.1 Bronchus and lung
162.2 Pleura
162.8 Multiple sites
163 Malignant neoplasm of lung, unspecified as to whether primary or secondary
OSWER has failed to subtract the mesothelioma deaths to get the number of lung cancer cases.
Using linear regression, I calculate a different value for the intercept (116), same as Enterline's.
From Poisson regression log(Obs/Exp) = 0.523 at CumExp = 0. ==> intercept is 1.68. Neither of these is
the same as the alpha displayed on Figure A3-4.
Uncertainty in Particle Size data: Lumping all jobs, and factories, together is suspect.
A4. Ontario Asbestos cement
There is a mesothelioma exposure-response study:
Finkelstein M. The exposure-response relationship for mesothelioma among asbestos-
cement factory workers. Toxicology and Industrial Health 1990;6:623-627. , and
Finkelstein M. Analysis of the exposure-response relationship for mesothelioma among
asbestos-cement factory workers. Ann NY Acad Sci 1991;643:85-89.
A5: New Orleans Cement Products Manufacturing Plants
Uncertainty in Particle Size Distributions
Chrysotile: Averaging over jobs and processes
Amphibole: Combined mining/milling and insulation data sets
Discussion:
Major difficulty is with particle size distributions. All jobs and processes assigned similar
distributions. Smearing over distributions averages out all distinctions and is noninformative.
A6: Quebec Mines and Mills
Conversion Factor:
Gibbs (1994): "there were definite patterns with Membrane Filter/Midget Impinger ratios
increasing from the predominantly dust-generating operations such as drilling and crushing to fibre-
releasing operations such as fibre screening and bagging."
3-30
-------
==> substantial uncertainty in conversion factor
Cumulative Exposure:
OSWER states that the "study reports CE rather than CE10. This is not correct. The study reports CE to
age 55. This introduces comparability issues with studies reporting CE or CE10.
SMRs are calculated from age 55 onwards.
Uncertainty in Particle Size Distributions
Chrysotile: Averaged over mining and bagging. Uncertainty stated to be low, but differences are
smeared out.
Amphibole: Data from talc mining and milling applied to chrysotile mining and milling. Needs to be
demonstrated that this is reasonable.
Discussion:
Major difficulty is with particle size distributions. All jobs and processes assigned similar. Smearing
over distributions averages out all distinctions and is noninformative. Needs to be demonstrated that
using tremolite size distribution from talc mining is relevant for Quebec chrysotile mining.
A7: Pennsylvania Textile Factory
ICD Coding of Malignant Neoplasms
Death certificates obtained and coded by nosologist. The nosologist coded to ICD 7:
162 Malignant neoplasm of bronchus and trachea, and of lung specified as primary
162.0 Trachea
162.1 Bronchus and lung
162.2 Pleura
162.8 Multiple sites
163 Malignant neoplasm of lung, unspecified as to whether primary or secondary
164 Malignant neoplasm of mediastinum
McDonald et al classified lung cancer as ICD 162-164
In ICD 7, mesothelioma and lung cancer both coded to ICD 162. The authors state that
there were 10 pleural tumours.
Lung Cancer Results:
The author of the OSWER report ignores the fact that Table 5 in the McDonald report
(FigureA7-l) combined lung cancer + mesothelioma + ICD 160 + ICD 161
3-31
-------
Mesothelioma:
Because of the combination of endpoints, many of these meso deaths were included with the lung cancer
analysis above.
The computations to produce a meso risk estimate are tenuous.
Bias Correction Factor CE vs CE10
McDonald et al present table 5 in relation to dust exposure accumulated to 10 years before death. They
do not state how they dealt with subjects who were still alive.
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences.
Discussion
OWSER has failed to recognize that mesothelioma deaths have been combined with lung cancer and
other respiratory sites in Table 5. The numbers of observed lung cancer deaths are thus in error.
A8: Connecticut Friction Products
Cumulative Exposure Estimates
Estimates made for Departments rather than jobs
Mesothelioma Data
No cases observed
OWSER make approximations and assumptions about PYR and cumulative exposure.
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences.
A9: British Textile Factory
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences.
A10: Italian Chrysotile Mine
Fraction Amphibole: OSWER guesses that amphibole was present despite the authors report that none
was present and that a fibrous silicate, balangeroite, was present.
3-32
-------
Lung Cancer Results: It is not clear from the published reports whether or not PYR were
assigned to dust categories sequentially as work time accumulated, or whether all PYR were attributed
to final category. Failure to assign PYR sequentially would flatten dose-response curve.
Mesothelioma Results: OSWER estimates duration and exposure data. The Piolatto data are superceded
by those of Mirabelli et al 2008.
(Mirabelli D, Calisti R, Barone Adesi F, Fornero E, Merletti F, Magnani C.
Excess of Mesotheliomas after Exposure to Chrysotile in Balangero, Italy.
Occup Environ Med. 2008 Jun 4. )
Uncertainty in Cumulative Exposure for Meso: Not certain whether authors moved PYR
through exposure categories or assigned all PYR to final one achieved
Uncertainty in Fraction Amphibole: There is no tremolite, but another fibrous silicate in the ore. This
was said to be 0.2 - 0.5% of the mass of chrysotile.
Applying size data for tremolite is inappropriate.
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences.
A12: Swedish Cement Plant
Relative Risk: The OSWER author treats this study as a standard SMR-type study, which it is not. There
was a defined industrial reference population of 1233 men, not a national reference population. Relative
risk was computed by age and calendar period adjusted Poisson regression modeling. The data presented
in Figure A12-1 is thus not meaningful. There is no measure of "expected deaths" for the referents, and
where the author derives a RR of 1.8 for the referents is a mystery. He has apparently misinterpreted
Table 2. Allocation of Observed deaths based upon person-years is not necessarily valid as the relative
risk is a regression-adjusted estimate.
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences.
The assignment of particle size data for the amphiboles is highly speculative.
A13: Libby Montana Vermiculite Mine
3-33
-------
Pg A13-4. OSWER states that exposure is expressed in terms of f/cc lagged by 10 yrs. This is not true.
There is no lag.
Relative Risk: The OSWER author treats this study as a standard SMR-type study, which it is not.
Comparisons were internal to the cohort, not external to a national reference population. Relative risk
was computed by adjusted Poisson regression modeling. The data presented in Figure A12-1 is thus not
meaningful. There is no measure of "expected deaths" for the referents. There is no 95% confidence
limit for the Reference Category and no value of a.
A14: Wittenoom Australia Crocidolite Miners
OSWER fails to recognize that this is a case-control study. Figure A14-1 is thus not meaningful. There
is no confidence interval around the Odds Ratio for the reference category. The concept of observed and
expected is not well defined. The 90% confidence Intervals are probably inaccurate.
OWSER estimates cumulative exposure by multiplying concentration by estimate of average duration.
This is bound to introduce misclassification.
Uncertainty in Particle Size Data
Again OSWER averages over departments and jobs, smearing out any differences. The statement that
the uncertainty in fs;ze is low is not very credible.
A15: Belgian Asbestos Cement Factory
Issues:
Cause of death not based on official records, but from family doctor or social workers who visited
relatives. This is unreliable.
Exposure concentration: Fibre counts available 1970-76. Exposures estimated 1928-77 using a logistic
function. Would expect levels to be better modeled by step-function with changes occurring with
ventilation, layout, or work practice changes.
Recommendation: This study should be omitted because the mortality data for the cases (family doctor
or social workers) are derived from a different source than the reference data (death certificates).
A16: Austrian Cement Factory
Lung Cancer SMRs. Figure 16-1 uses smoking-adjusted SMRs. This is one of the few studies where this
desirable adjustment is possible, but the use of smoking adjusted SMRs is not consistent with the SMRs
abstracted from the other studies in the database.
Uncertainty in Particle Size Data
3-34
-------
Again OSWER averages over departments and jobs, smearing out any differences. The statement that
the uncertainty in fs;ze is low is not very credible.
A17: China Asbestos Products Factory
Relative Risk: The OSWER author treats this study as a standard SMR-type study, which it is not.
Comparisons were internal to the cohort, not external to a national reference population. Relative risk
was computed by Cox proportional hazards regression modeling. The data presented in Figure A17-1
are thus not meaningful. There is no measure of "expected deaths" for the referents. There is no 95%
confidence limit for the Reference Category and no value of a.
Dr. Andrew Gelman
Below are my responses to the charge questions. Before answering these, let me emphasize that I am
not an asbestos expert and, although I have looked at the EPA report ("Proposed approach..."), I have
not looked at the original research that is cited in the report. Also, as you can see, I've focused my
comments in the areas where I am more expert.
1. Yes, I agree that the data indicate that differences in effects of different sorts of asbestos may exist
and that an effort of this type is warranted.
2. The scientific basis for the approach seems reasonable to me. There are certainly challenges here in
combining data from studies that have different sorts of measurements. In general, I'd recommend
including more categories rather than fewer. I can't comment on the biomedical models of health
effects.
3a. The mixture model approach makes sense to me. I cannot really comment one way or the other on
the specific risk models that are being used (the models of risk given exposure and how this is affected
by the time of exposure).
3b. I don't have any particular suggestions of other model forms to be investigated; however, if other
reasonable models are proposed, I agree that they should be looked into.
3c. Interactions of smoking and other cancer risks can be large. For example, I know that the added
cancer risk from radon exposure is much higher among smokers than nonsmokers. If the data are
available, it would make sense to fit separate models for smokers and nonsmokers. Otherwise it might
make sense to use a model in which the added risk from asbestos is higher (by some multiple) for
smokers than for nonsmokers.
4a. The recommendation in Section 8.3 appears to be to fit separate rates for each group in each study.
This makes sense to me. As noted in Section 8.3, this allows more direct modeling of the data. Perhaps
this approach has a disadvantage if it is not easy to replicate the methods that were used in preparing the
derived statistics for each study.
4b. The Poisson likelihood might be reasonable; however, in practice we almost always use
overdispersed models. If the Poisson model is indeed used, it is important to check for overdispersion in
3-35
-------
the differences between data and fitted models. In some settings, there is a group-level variance
parameter which catches the extra-Poisson variation and eliminates the need for an overdispersed model.
That may be happening here (through the alpha_s and Q_sg parameters in the model in Section 8.6.1)
but I am not completely sure.
5a. Section 8.4 discusses many potential sources of uncertainty. I do not know of other sources that
should be accounted for, but, again, I am not an expert here.
5b. Characterizing the uncertainties as independent probability densities can't be correct, but perhaps it
is a reasonable thing to do in practice. The use of "professional judgment" is problematic because such
judgments are generally found to be overconfident (that is, often not containing the true values);
however it's not clear what the alternative is. It could be useful to propagate each source of uncertainty
to see how it contributes to the uncertainty in the final recommendations.
5c. The general approach for characterizing uncertainty used in Appendix C seems reasonable, but,
again, with all the details, I could imagine that something important could have been left out.
5d. The assumption that errors combine in a multiplicative fashion could be reasonable. I am not sure
how this assumption could be tested given the available data.
6a. I'm not sure why "measurement error models" are labeled as "weighted regression models." My
impression is that measurement error models are simultaneous equation models (e.g., a regression model
for y given x, along with a measurement model for x.observed given x, where in both cases, x is the true
predictor and x.observed is the predictor measured with error). I don't see where "weighted regression
comes in."
6b. I don't think weighted likelihood analyses are helpful here. For one thing, there's no direct way to
get standard errors from weighted likelihood analysis; for another, the weights should, to be correct,
include model error as well as measurement error, so that the appropriate weights depend on estimated
parameters such as group-level variances.
6c. I think the Bayesian approach is most appropriate.
7. I don't fully understand the prior distributions in Section 8.6.2. It says that the alphas are likely to
fall between 0.5 and 2, but then it puts a Uniform[0.1, 10] prior distribution on each. I have two
suggestions:
(a) Check the inferences for the parameters after fitting the model. If the estimates are far from the
originally suggested range of (0.5, 2), then see what's going on. Is it just a matter of there being a large
posterior uncertainty? If so, perhaps more prior information would be useful. Or are the alphas
estimated with precision to be far from their prior range (e.g., an estimate such as 4.0 with posterior sd
of 0.5)? If the latter, you have to think harder about what is going on here. I have a similar comment
for the priors on the KL_b and KM_b parameters.
(b) Consider a hierarchical model for these parameters. If you have many alpha_s parameters, you can
give them a prior dist with parameters estimated from the data. That could make more sense than giving
them independent noninformative distributions.
3-36
-------
8. The bins seem to be chosen in a somewhat ad hoc, data-based manner. This is fine, but I'd suggest
including more bins rather than fewer. I don't see the advantage in reducing the number of bins. I
mean, I can see why you wouldn't want hundreds of bins, but I'd think that 5 or 10 bins wouldn't be a
problem.
9. Again, I wouldn't focus so much on the binning strategy. I'd choose a reasonably large number of
bins that could include whatever variation might be expected, then fit the model from there. I could see
going back to the binning question at the end, when you get to policy questions, but I think it makes
more sense to estimate for many bins and then do some smoothing at the end, rather than trying to
combine bins in the main analysis.
Beyond this, I'd recommend posterior predictive checks, i.e., simulating replicated datasets from the
model and seeing if they look like actual data. To do this, it would be first necessary to create some
useful graphical displays of the data that are being used to fit the model. Analysis of residuals is fine,
but it would be useful to also display raw data and replications of the raw data.
10. The idea of sensitivity analysis is fine although in practice I don't know that much is learned from
these things. Still, I think it has to be done.
11. I don't really have any comments on the rules for selecting studies in the meta-analysis. I defer to
others who are more familiar with the biomedical literature.
12.1 defer to others on whether there are other relevant studies of lung cancer or mesothelioma that
should be considered.
13. The extrapolation from dust to asbestos might be questionable but I don't know that it can be
avoided. You could see what happens if these studies are excluded entirely, but if that leaves inferences
that are too vague, maybe you have to go with the assumption.
14. I can't judge the validity of the assumptions underlying the exposure interpolations, but it seems
reasonable to use the published data if that is the closest thing available.
15. Given the methods that are being used, the uncertainty about lifetime cancer risks can be
summarized using simulations. However, I would also recommend estimating life-years lost, not just
lifetime cancer risks. That way, costs can be expressed as dollars per life-year, which is a slightly
different measure than dollars per life. Also could be broken down as estimated lives and life-years lost
among smokers and among non-smokers.
Mr. John Harris
General response
The microscopic techniques used by EPA for risk evaluations have both benefits and limitations. The
chief advantage to microscopy is to supply important size data information used for binning purposes.
There are some modifications to these methods that critical information for risk assessment researchers.
3-37
-------
PCM has historically been criticized for not providing any information of the fiber types counted. For
mineral identification of asbestos in bulk materials, polarized light microscopy (PLM) has been the
standard. However, for smaller fibers, PLM is limited since the analysis is done at lOOx as compared to
400x by PCM. Recently, there are new innovations that allow modified PCM microscopes to utilize the
mineral identification techniques of PLM on a PCM microscope. This can be a useful tool for
investigators that have historical PCM slides that could be recounted to determine chrysotile:amphibole
ratios at the PCM level. For studies with limited funding, this might be a good alternative approach to
standard PCM alone.
TEM is the best approach for studies needing definitive analysis as described in the documentation
provided for this advisory board. Typically for risk assessments, the ISO method (ISO 10312) is
preferred as it offers the most analytical intensive method to define and count all fibrous particles
regardless of size. The counting rules can be modified to allow analysts to include all fibers that would
be normally counted by PCM, or PCM equivalent sized fibers.
One drawback to TEM is the reduced amount of area analyzed. Most asbestos fiber size distributions
typically have very few long, thin fibers compared to a very high number of shorter fiber sizes. The
reduced analytical area covered by TEM limits the detection of longer, thinner fibers under these
conditions. Therefore, a stratified analysis approach makes better sense for TEM studies for historical
or current risk assessments. The stratified analysis would use both lower magnification to search only
for longer fiber sizes and higher magnification to analysis for all fiber sizes. The analyzed area at low
magnification would be relatively large while the analyzed area for high magnification would be
standardized to an established sensitivity required.
TEM studies provide a wealth of information of each mineral type present. TEM identification of
amphiboles requires more rigorous mineralogical identification using the International Mineralogical
Association (IMA) protocol proposed by Leake (1997). TEM chemical systems must be calibrated to
better standards, such as microprobe standards, in order to more accurately identify differences in
mineral types, especially with amphibole minerals.
In addition to having the correct chemical identification of the mineral type, accurate diffraction data is
needed to differentiate minerals with similar chemical compositions to the mineral types of interest. Use
of internal aperture standards is not sufficient. Measuring diffraction spacings over several rows or
diffraction points are needed to accurately measure crystallographic dimensions to differentiate different
mineral classes. This requires more astute attention to details for TEM analysts than routine samples
containing commercial grade asbestos.
Some of the ISO counting rules may affect fiber dimensions and bias fiber dimensions for binning. For
example, fibers with one end sufficiently embedded in a particle are given an estimated fiber length
based primarily on the size of the particle and not the fiber itself. Fibers intersecting with TEM grid
bars are recorded as twice their visible length. Complex structures with more than 9 individual fibers
are noted as "+" instead of allowing an analyst to measure and count all structures. For complex
structures with more than 5 substructures, the additional structures after the first 5 are recorded as
"residual" and are estimated based on an average length and width.
3-38
-------
Using a TEM combined with SEM capabilities (STEM) solves most of these fiber dimension problems.
This enhanced capability allows an analyst to determine whether a fiber is actually embedded in the
particle or lying free on the surface or under a particle. By looking at the surface features, we often see
round particles composed of asbestos fibers tightly bound that are not countable by any method. The
ability to analyze a complex fiber arrangement completely by diffraction, chemistry and surface features
provide an accurate description of that structure for binning and identification purposes. Using SEM
only for an evaluation of Superfund sites is problematic since only surface features and chemistry are
available. Without the higher resolution of TEM, higher penetration of electrons through the sample at
higher energies and diffraction capabilities, some particles may be identified definitively by SEM.
I hope this helps guide EPA in its efforts to accurately assess the environmental exposure at Superfund
sites.
Charge question 14a:
Are the point estimates and uncertainty distributions for the fraction amphibole term proposed for each
study scientifically valid?
Defer to statisticians, risk assessors.
Charge question 14b:
Is it scientifically valid to use surrogate TEM data to estimate bin-specific concentrations and exposure
values in studies where these data are not reported?
No comment
If not, what alternative approach could be followed , or what additional data would be helpful?
No comment
Charge question 14c:
Are there any additional bi-variate TEM data sets available that would be useful in this analysis?
Some additional variables could include properties such as surface area and mineralogy.
Charge question 14d:
Are the point estimates and uncertainty distributions for the fraction amphibole term scientifically
valid?
No comment
Charge question 14e:
Can you suggest any ways to improve the process used to identify select the best available matching
TEM data set(s) to a workplace?
Eliminate impinger data and PCM data as surrogate data for TEM. Use only TEM data sets only.
3-39
-------
How sensitive would the model output be to these changes?
Charge question 14f:
Would the model benefit by establishing a common lower cut-point in diameter to normalize the lower
detection limit across studies?
Charge question 14g
Do the studies included in the model have surrogate data of sufficient quality and similarity to expected
exposure conditions to support the model?
No comment
If not, what alternative approach could be followed?
Charge question 14h:
Are the PDFs described in Appendix C to characterize the uncertainty in the extrapolation ofTEM
particle size data from one location to another sufficient and helpful in understanding the implications
of the method used?
No comment
Charge question 14i:
Are the extrapolation techniques used on the raw TEMdata sets to meet bin definitions (e.g., 0.4 jum
diameter) transparent, objectively presented and scientifically valid?
No comment
Are there alternative techniques that you would recommend?
I would recommend that EPA either consider modifying ISO TEM methodology to better fit a more
accurate identification of particle types and sizes common to Superfund sites or develop TEM
methodologies of their own. Described below are some observations of our experiences with difficulties
encountered at some of these sites.
a. Current use of the ISO counting rules for PCM equivalency using TEM procedures can
create biases. These biases include:
i. ISO 10312 method does not include nonasbestos fibers in its counts. They are
noted as comments. Investigators should be aware of this bias when comparing
ISO data with PCM data.
ii. Matrices (fibers embedded in particulates) need surface imaging capabilities (i.e.,
STEM) to accurately determine the true fiber length as well as whether the fiber is
truly embedded or simply landed on the surface (see image below). For fibers
lying underneath particles, the TEM grid can be turned over and the fiber
visualized on the bottom surface of the replica. There may be possible methods
to determine fiber length within a matrix if needed. Under ISO counting rules,
3-40
-------
this structure length would be measured by the analyst at more twice the true
length since more than 30% of the virtual length of the protruding fiber is
obscured. By increasing a fiber length, the aspect ratio of the fiber is also
increased and incorrectly reported.
in.
For investigations of activity-based sampling involving risk assessments, our
laboratories looks for noncountable particles with compacted fibrous asbestos that
would not be included in either PCM counts nor ISO counting rules. These types
of unique asbestos-laden particles need to be included in counts until future
investigations determine the health effects of these types of particles. Shown
below is an example of one of these particles from a naturally-occurring asbestos
site in Swift Creek, Washington:
IV.
Structures touching grid bars are given a measurement of twice its visible length.
Most often, this creates a longer structure than actually exists.
3-41
-------
v. There are limitations in recording data for complex structures that may bias
reported numbers of fibers detected by PCM but not recorded effectively in ISO
method studies. Substructure (or "total structure") counts larger than 10 are given
a "+" and are no indication of the actual fiber loading. When more than 5 total
structures are part of a primary structure, a designation of "residual" is applied
and total structure counts are estimated. All total structures should be included.
For example, the chrysotile structure below from the Atlas mine in California
shows a complex structure that would be defined primarily using a "+"
designation instead of detailed numerical values.
1 : 531
10m
vi. All structures should be digitally imaged for future investigators as binning
categories and other morphological characteristics are modified.
2. TEM EDS chemical analysis of amphiboles need to be standardized based on mineralogical
methods such as the International Mineralogical Association method by Leake from 1997
There is no standardization among asbestos labs providing analysis which can create biases
between PCM and TEM results.
The determination of asbestiform or nonasbestiform varieties of minerals should not be done during the
analysis by the analyst. Instead, it must be done post-analysis by epidemiologists. Since the definition
of cleavage fragments has been and will continue to be a moving target, all potential structures that fit
3:1 aspect ratio with minimum fiber lengths should be included in counts for future binning to determine
suitability for current and future studies.
3-42
-------
Dr. Karl Kehey
1) Do you agree that the data are sufficient to indicate that such differences may exist and that an effort
of this type is warranted?
The committee chose to rephrase this question. I believe this to be a wise choice, as the question, as
posed is confusing. I would agree that the epidemiologic data are consistent with the hypothesis that
asbestos mineral type and particle dimensions have different potency for mesothelioma. This is based
upon the multiple, consistent epidemiologic observations demonstrating differences in the occurrence of
MPM in workers whose exposure was known to be to different types of asbestos.
However, I do not agree that the data are sufficient to indicate that differences in potency by mineral
type or dimension exist in the case of lung cancer. The data are not consistent for lung cancer, with the
South Carolina workers particularly standing out as a group primarily exposed to crysotile with very
dramatic lung cancer risks.
In sum, the data are coherent for the suggestion that fiber type and dimension are important for MPM
I am (and I believe the committee was) also supportive of an effort to represent this scientific consensus
in the risk assessment process. The approach that is proposed, however, is flawed. The flaws are
potentially fatal ones. Chief among the flaws is the attempt to use the occupational exposure data.
These data are sparse and truly not amenable, in the vast majority of cases, to the multiple binning
approach being proposed. Even the two bin approach is problematic, based upon the nature of the
exposure data.
Finally, and importantly, the document presented for review by the committee was profoundly
inadequate with respect to its representation of the animal and mechanistic data. Indeed, for example,
there are data that suggest that there may be heritable susceptibility to the action of asbestos in
generating MPM (see multiple publications from Carbone,.including: Dogan AU, Baris YI, Dogan M,
Emri S, Steele I, Elmishad AG, Carbone M; Genetic predisposition to fiber carcinogenesis causes a
mesothelioma epidemic in Turkey. Cancer Res. 2006 May 15;66(10):5063-8 and Roushdy-Hammady I,
Siegel J, Emri S, Testa JR, Carbone M. Genetic-susceptibility factor and malignant mesothelioma in the
Cappadocian region of Turkey. Lancet. 2001 Feb 10;357(9254):444-5). That is, there may be a
susceptible subgroup of the population and, if that is the case, assuming that this subgroup responds to
asbestos in the same fashion as the population as a whole is not conservative.
The work by the EPA, while laudable in its intent, is flawed in its execution. I am supportive of going
forward with investigation of the impact of fiber type and dimension on mesothelioma risk, but any
modeling exercise must be clearly labeled as what it is - the document reviewed has only the very
loosest of biologic ties, does not consider susceptibility, includes uncertainty in exposure assessment
that cannot conceivably be scientifically estimated and, in essence, seeks only to fit epidemiologic data
to a better underlying mathematical construct. This effort is well intended and is an earnest and
important research enterprise but it is not supportable as an endeavor that will contribute to better
estimation of disease risk; that is, it is not an effort that is in the interest of overall public health at this
time.
2) Please comment on the adequacy of the overview of EPA 's OSWER Draft Report.
The sections of the report that serve as the scientific basis for the proposed dose-response modeling are
very poor. As noted in my response to charge question 1,1 would advocate that this document
3-43
-------
acknowledge its tenuous relationship to biologic mechanism and not include the mechanistic sections.
They are manifestly inadequate as written.
Simple examples: nowhere in this document is there any reference to the now large body of data that
suggests that there may be a portion of the population that is susceptible to the toxic effects of asbestos
(Carbone, M - multiple references above). The data describing the toxicology and "mode of action" of
asbestos ignores completely the fact that both lung cancer and mesothelioma are known to harbor
significant epigenetic gene alterations that contribute to the genesis of these diseases (Suzuki M,
Toyooka S, Shivapurkar N, Shigematsu H, Miyajima K, Takahashi T, Stastny V, Zern AL, Fujisawa T,
Pass HI, Carbone M, Gazdar AF. Aberrant methylation profile of human malignant mesotheliomas and
its relationship to SV40 infection. Oncogene. 2005 Feb 10;24(7): 1302-8.; Christensen BC. Godleski JJ.
Marsit CJ, Houseman EA, Lopez-Fagundo CY, Longacker JL, Bueno R, Sugarbaker DJ, Nelson HH,
Kelsey KT. Asbestos exposure predicts cell cycle control gene promoter methylation in pleural
mesothelioma.Carcinogenesis. 2008 Feb 28.;Destro A, Ceresoli GL, Baryshnikova E, Garassino I,
Zucali PA, De Vincenzo F, Bianchi P, Morenghi E, Testori A, Alloisio M, Santoro A, Roncalli M.Gene
methylation in pleural mesothelioma: correlations with clinico-pathological features and patient's
follow-up.Lung Cancer. 2008 Mar:59(3):369-76.:Tsou JA. Galler JS. Wali A. Ye W. Siegmund KD.
Groshen S. Laird PW. Turla S. Koss MN. Pass HI. Laird-Offringa IA.DNA methylation profile of 28
potential marker loci in malignant mesothelioma.Lung Cancer. 2007 Nov;58(2):220-30.
Pu RT. Sheng ZM. Michael CW. Rhode MG. Clark DP. O'Leary TJ.Methylation profiling of
mesothelioma using real-time methylation-specific PCR: a pilot study.
Diagn Cytopathol. 2007 Aug;35(8):498-502.)
There are also omissions of major animal works germane to the question of how fiber size and type
impact mesothelioma and lung cancer risk.
3a) Do you agree that the lung cancer and mesothelioma risk models that are proposed are a
scientifically valid basis for this fitting effort?
This is a complex question. The models are an attempt to parse crude data in a parsimonious fashion
consistent with a set of assumptions. Uncertainly in exposure makes testing for validity impossible.
Fitting the data are possible, but this does not necessarily imply the models are valid.
3b) Should additional model forms be investigated?
Yes; Dr. Peto had several suggestions that might be profitably pursued.
3c) For lung cancer, the current risk model is multiplicative with the risk from smoking and other cause
of lung cancer. Should the nature of the interaction between asbestos and smoking be investigated
further?
Given that there is no consensus on the mode of action of asbestos in generating the multiplicative
interaction with tobacco, additional investigation is most certainly needed. The epidemiology strongly
suggests that the interaction is multiplicative, but awaits data that might allow for a more complete
description of the biology and the nature of this association.
3-44
-------
If so, how should this be done?
The research community might be asked to investigate the nature of this interaction in grant funded
research with and RFA that precisely articulated the questions of interest.
Do you think the model would be sensitive to additional quantification of the interaction between
smoking and asbestos?
Yes. Any synergy will skew models and be very sensitive to this interaction in a modeled dose-response.
4a) Is fitting at the group level (based on the number of cancer cases observed) preferred to fitting at
the study level (based on the study-specific KL or KM values)?
Fitting should be done in a fashion that is least prone to propagate error and misclassification.
What are the advantages and disadvantages of this approach?
As above.
4b) If so, is it scientifically justifiable to use a Poisson likelihood model for the observed number of
cases in each group?
Poisson models seem appropriate for sparse data.
8a) Do you agree that multiple binning strategies should be evaluated, or do you believe that a
physiological basis exists that can be used to identify a particular set of length and width cutoffs that
should be assessed?
I believe that multiple binning strategies are an attractive intellectual advance in risk assessment.
However, there is really only one study that has compared anywhere near the number of exposure
assessment techniques that would be needed to accurately assess the effects of using a multiple binning
strategy. On the face of it, I believe that this is not possible. It will amplify uncertainty in unknown
ways and is simply not good science.
If so, what would these length and width cutoffs be, andean these bins be implemented considering the
limitations in the available TEMparticle size data sets?
The human data primarily address below and above 10 um only. The data cannot address width
adequately, as prior methods were unable to assess small width fibers. Hence, this cannot be done with
any scientific validity.
8b) Are there any of these strategies that you feel do not warrant evaluation?
Yes.
3-45
-------
If so, why?
The binning approach is not feasible as it propagates uncertainty, is based upon selective and incomplete
data, has not been subjected to sensitivity analysis, and does not consider susceptibility.
8c) Assuming that fitting is performed using Bayes-MCMC, OSWER is proposing that a comparison of
goodness of fit between different strategies be based on the Bayes Factor.
Do you agree that this is a statistically valid method for comparing binning strategies?
NA
Are there any other comparison methods you would recommend?
No.
8d) Is it important to account for differences in the number of fitting parameters (bin-specific potency
factors) when comparing 1-bin, 2-bin, and 4-bin strategies to each other?
NA.
If so, how should that be done?
Dr. PaulLioy
Section 8.4 - Characterizing Uncertainty in Exposure Data
5 a. Have all of the important sources of uncertainty in cumulative exposure matrices been identified? If
not, what other sources should be accounted for?
Based upon the observations of Dr. Peto, no. However, the most pressing issues relate to the need of a
defined research program, especially for bullets 7-9. Experimental data is required to provide values
other than default factors for long term applications of the Bin approach to Asbestos at Superfund sites
and other fibers. For Asbestos alone, at superfund sites, the ratio of dust to fibers will be variable and
not all associated with the asbestos fibers. Thus the need for data on actual fiber counts in the selected
Bins, or any other Bins.
5b. Is it appropriate to characterize the uncertainty from each source in terms of an independent
probability density estimated using professional judgment? If not, what alternative approach is
suggested?
Yes, this is fine for the initial application, but not necessarily reality. For the most important issues you
will need better designed studies, and appropriate data collected to minimize uncertainties. You will be
hard pressed to make totally justifiable assumptions, thus the need to do research at this time.
3-46
-------
5c. Are the general strategies for selecting distributional forms and parameter values described in
Appendix C (and applied in Appendix A) appropriate for characterizing uncertainty in exposure
matrices? If not, what alternative strategies are recommended?
Yes, comprehensive.
5d. Based on the assumption that each of the sources of error is independent, OSWER is proposing an
approach where the errors combine in a multiplicative fashion. Please comment on the scientific
validity of this approach and provide detailed suggestions for other approaches OSWER should
consider.
For uniform processes the error can be assumed to be independent, but it is not necessarily true for
asbestos at superfund sites, which is a mixed waste from disparate origins.
Sum of absolute values of errors is more conservative, as well as the root mean square of the
uncertainties.
Section 8.5. Fitting Approach
6a. Is it appropriate to account for measurement error in the exposure data by using "measurement
error" models (weighted regression methods)? If so, how would the weights assigned to each exposure
value be assigned?
Yes, weighing is reasonable but should be conditional based upon those variables having the highest
impact on the estimates of exposure.
6b. Is the assignment of a PDF for data quality sufficient or should data quality be factored into a
weighted likelihood analysis?
No comment.
6c. Do you think that the proposed strategy of fitting the risk models to the available epidemiological
data using Bayes-MCMC is scientifically justifiable? If not, what alternative strategy do you suggest,
and why?
Yes. An excellent application for a modeling procedure that will improve results and evolve
understanding over time based upon new data. Based upon the discussions at the meeting, new data is
essential since dust to PCM ratios from industrial and occupational studies are not directly applicable.
Section 8.62 - Specification of Priors
7. Do you think that the proposed strategy of fitting the risk models to the available epidemiological
data using Bayes-MCMC is scientifically justifiable? If not, what alternative strategy do you suggest,
and why?
The Bayes approach requires appropriate probability density functions. Given the level of understanding
a uniform distribution with wide boundaries is a reasonable start for sensitivity analysis. These need to
3-47
-------
be augmented with more appropriate data on asbestos at superfund sites to achieve a normal or log
normal PDF. The Agency should take advantage of all the work being funded at Libby to begin to
validate the Bin approach. This is an opportunity that should not be lost!
Section 10.2 - Extrapolation from Dust to PCM-Based Measures
13a. Is it scientifically justifiable to employ a default dust-to-P CM conversion factor when there are not
site-specific data available?
After careful consideration of the extensive discussions made by fellow committee members at the
meeting, No, it is not justified to use a default. The basic premise of transferability of dust in an
industrial setting to superfund application has too much uncertainty. This problem needs to be
eliminated by better data and determining the strength and utility of TEM measurements in the overall
"bin" approach.
The agency needs to use the South Carolina data as a first level attempt at understanding of the utility of
the Bin approach. However, it must look for opportunities, like Libby, to establish a comprehensive
framework for the future validation of the Bin approach.
13b. Are the uncertainty distributions specified in Appendix A to characterize the uncertainty in this
extrapolation consistent with available information and are they statistically appropriate?
Yes.
Section 10.3 Extrapolation from PCM to Bin-Specific Measures
14a. Are the point estimates and uncertainty distributions for the fraction amphibole term proposed for
each study scientifically valid?
No, uncertainties in the measurement and estimation techniques. Again the Libby research and the
South Carolina data must be fully exploited in validation of the Bin approach.
14b. Is it scientifically valid to use surrogate TEM data to estimate bin-specific concentrations and
exposure values in studies where these data are not reported? If not, what alternative approach could
be followed, or what additional data would be helpful?
I have no answer, except what are the biases that are introduced, and how does one reasonably select the
error PDF without data? The analyses will be highly uncertain, and could lead to poor conclusions. You
need to obtain the TEM data to reduce uncertainties. Sort of an obvious answer.
14c. Are there any additional bi-variate TEM data sets available that would be useful in this analysis?
No answer, I am not knowledgeable on this point.
14d. Are the point estimates and uncertainty distributions for the fraction amphibole term scientifically
valid?
3-48
-------
See answer to 14a.
14e. Can you suggest any ways to improve the process used to identify select the best available
matching TEMdata set(s) to a workplace? How sensitive would the model output be to these changes?
No. However, this is related to 14b.
14f. Would the model benefit by establishing a common lower cut-point in diameter to normalize the
lower detection limit across studies?
Yes, but it should be based upon inter-comparison studies of blind samples among the TEM analysts.
14g. Do the studies included in the model have surrogate data of sufficient quality and similarity to
expected exposure conditions to support the model? If not, what alternative approach could be
followed?
No, need to obtain data that can be of value in validation studies
14h. Are the PDFs described in Appendix C to characterize the uncertainty in the extrapolation of TEM
particle size data from one location to another sufficient and helpful in understanding the implications
of the method used?
Yes, because of the ability to complete sensitivity analyses.
141. Are the extrapolation techniques used on the raw TEMdata sets to meet the bin definitions (e.g. 0.4
jum diameter) transparent, objectively presented and scientifically valid? Are there alternative
techniques that you would recommend?
Yes.
Section 11 Utilizing Potency Factors to Compute Life Time Risk
15a. What method is best for estimating the uncertainty in lifetime cancer risk prediction?
A validation study must be completed to get a much better handle on uncertainties, and the overall
utility of the bin method.
15b. Assuming that estimates of exposure at Super fund sites will also have uncertainty, how should the
overall uncertainty in risk predictions be characterized?
Ambiguous question. For all risks or just asbestos risk? Site specific applications and data are usually
the best way to minimize uncertainties.
3-49
-------
Note, as discussed at the meeting:
The agency needs to do a better job of explaining the purpose of the proposed appraoch, initially
superfund sites, and any intention of applying it to other asbestos issues, and fibers. As written, the
committee was not given a full understanding of the over all conceptual framework
General Comment:
I agree with Dr. Lippmann on the need to have the Bin approach generalized for all fibers. In the US,
Asbestos is primarily a Superfund and removal issue, however, other fibers are in use and being
produced. Note the aftermath of the WTC did not have any reasonable standards for clean up of many
vitreous fibers etc. , and we still do not in 2008. I cannot re-emphasize this point more.
Dr. Mort Lippmann
Overview Remarks
While the document being reviewed is narrowly focused on OSWER needs, any endorsement will lead
to a document that has much wider implications to risk assessments for airborne fibers within and
beyond EPA. Thus, it is important that the document be based on a broader review of the effects of
fibers on human diseases, and of the properties of fibers that affect these diseases. In order to provide
this perspective, the document should:
• Discuss fiber properties (lengths, widths, and biopersistance), and whether they differ as they
influence mesothelioma, and lung cancer in humans and rats (Rodelsperger 2004).
• Discuss lessons learned from long-term animal inhalation studies to asbestos, other mineral
fibers (Wagner et al. 1990), and synthetic vitreous fibers (SVFs) (Miller et al., 1999,), as well as
those learned from human experience.
• Discuss biopersistence data from animal inhalation studies involving synthetic vitreous fibers
(SVFs) with varying in vitro solubilities and in vivo solubilities in the lung as they inform
differential cancer risks of chrysotile and amphibole asbestos fibers (Eastes and Hadley 1995,
1996), and commercial chrysotle asbestos (Davis et al. 1985) and chrysotile asbestos that is
contaminated with tremolite asbestos (Pooley and Wagner 1988).
• Discuss relevance of fiber dimensions and biopersitence to risks from exposure to multi-walled
carbon nanotubes. The Poland et al. (2008) paper demonstrates that nanotube bundles with >24%
of lengths >15 urn were as or more toxic in short term assays as the long asbestos fibers shown
by Davis et al. (1985) to be more carcinogenic than UICC amosite, while amosite with virtually
no fibers <5 z/m in length produced no tumors..
The risks of exposure to chrysotile fibers should be considered separately from those of amphibole
fibers. Both are important, but they are not the same, even when fiber length and width are considered. It
is important to consider both, because:
• Amphibole fibers in Libby, MT are of immediate concern, and can best be addressed by
considering analogous human risks associated with other amphibole exposures. Also, a focus on
exposures to fibers from the Libby mines and the human experience from those exposures is
warranted because there are membrane filters that can be analyzed by modern measurement
methods that can provide the most comprehensive data set on the lengths, widths, and
compositions of amphibole fibers associated with a significant number of cancer cases that were
well diagnosed.
3-50
-------
• Exposures to chrysotile asbestos remain the most common exposures to carcinogenic fibers in
the US, and EPA's guidance is needed on the risks associated with building maintenance and
demolition.
• Much of the chrysotile asbestos in-place in buildings came from the Thetford mines in Quebec
or other mining regions where a significant fraction of the raw fibrous material was tremolite
asbestos, an amphibole form of asbestos that is much more biopersistent than chrysotile. The
risks are therefore greater than those from amphibole-free chrysotle.
The concept of binning by fiber composition (e.g., amphiboles vs. chrysotile) and by fiber
dimensions (e.g., length and/or width intervals) is sound and desirable in terms of the use of descriptors
of measurable exposure variables that are much more closely related to the health risks associated with
fiber inhalation. While the draft that we reviewed did not provide a sufficient amount of data on fiber
size to justify any specific size-related bins, the SAB Panel's discussions pointed the way to data
resources that would enable the authors to do so in the next draft document, and we urge that they do so.
They must seize the opportunity to do so, because this area of the overall challenge is the one most
amenable to significant progress. The literature on the influence fiber size and composition on toxicity
and cancer is very extensive, albeit primarily in rodent models. However, in this case, interspecies
extrapolation has much less uncertainty than in most cases because the physiological and anatomical
differences affecting fiber deposition in, and clearance from, the airways are well known (Lippmann
1988, 1994, Case et al. 2000, Berman et al. 1995, Brody et al. 1981). Fortuitously, the recent
publication by Stayner et al. (2007) on the associations of long chrysotile fibers with lung cancer in
chrysotile textile workers in Charlestown, SC provides important new support for the importance of
long fibers in carcinogenesis that was first demonstrated by Stanton and Wrench (1972) in rat
instillation studies, and later by Davis and colleagues (1978, 1985, 1986, 1987) in a long series of 2-year
inhalation studies in rats. By contrast, the uncertainties inherent in the risk models are much greater,
with KLs and KMs varying among the historic studies by an order of magnitude or more, and with little
prospect that retrospective re-analyses can refine them further.
Finally, it is important that EPA gets the exposure-related risk issues right this time and there would
be no excuse if it doesn't, because:
• Our ability to routinely sample and identify airborne fiber distributions by length, width, and
chemical and crystallographic composition is much more mature than in 1986.
• Our understanding of the influence of fiber length, width, and biopersistence on fiber toxicity is
much more mature than in 1986.
• There would be continued reliance on outmoded risk models and grossly inappropriate air
monitoring methods, which would therefore continue to cause unnecessary public concerns
about de-minimus risks, unwarranted litigation, and asbestos risks that increase from
unwarranted removals of in-place asbestos.
References
Berman DW, Crump KS, Chatfield EJ, Davis JMG, and Jones, AD. 1995. The sizes, shapes, and
mineralogy of asbestos structures that induce lung tumors or mesothelioma in AF-HAN rats following
inhalation. Risk Anal. 15:181-195.
Brody AR, Hill LH, Adkins B Jr, and O'Connor RW. 1981. Chrysotile asbestos inhalation in rats:
3-51
-------
Deposition pattern and reaction of alveolar epithelium and pulmonary macrophages. Am Rev Respir Dis.
123:670-679.
Case BW, Dufresne A, McDonald AD, McDonald JC, and Sebastien P. 2000. Asbestos fiber type and
length in lungs of chrysotile textile and production workers: Fibers longer than 18|j,m. Inhal Toxicol.
12(suppl.3):411-418.
Davis JMG. 1987. Experimental data relating to the importance of fibre type, size, deposition,
dissolution and migration. In: Proceedings of 1987 Mineral Fiber Symposium, Lyons, IARC.
Davis JMG, Addison J, Bolton RE, Donaldson K, Jones AD, and Miller BG. 1985. Inhalation studies on
the effects of tremolite and brucite dust in rats. Carcinogenesis 6:667-674.
Davis JMG, Addison J, Bolton RE, Donaldson K, Jones AD, and Smith T. 1986. The pathogenicity of
long versus short fiber samples of amosite asbestos administered to rats by inhalation and intraperitoneal
injection. Br JExp Pathol. 67:415-430.
Davis JMG, Addison J, Mclntosh C, Miller BG, and Niven K. 1991. Variations in the carcinogenicity of
tremolite dust samples of differing morphology. Ann NYAcad Sci 643:473-490.
Davis JMG, Beckett ST, Bolton RE, Collings P, and Middleton AP. 1978. Mass and number of fibers in
the pathogenesis of asbestos-related lung disease in rats. Br J Cancer 37:673-688.
Eastes W, and Hadley JG. 1995. Dissolution of fibers inhaled by rats. Inhal Toxicol 7:179-196.
Eastes W, and Hadley JG. 1996. A mathematical model of fiber carcinogenicity and fibrosis in
inhalation and intraperitoneal experiments in rats. Inhal Toxicol. 8:323-343.
Lippmann M. 1988. Asbestos exposure indices. Environ. Res. 46:86-106
Lippmann M. 1994. Deposition and retention of fibres: Effects on incidence of lung cancer and
mesothelioma. Occup. Environ. Med. 51:793-798.
Miller BG, Jones AD, Searl A, Buchanan D, Cullen RT, Soutar SA, Davis JMG, Donaldson K. (1999).
Influence of characteristics of inhaled fibers on development of tumours in the rat lung. Ann Occup Hyg
43:167-179.
Poland CA, Duffm, R, Kinloch I, Maynard A, Wallace WAH, Seaton A, Stone V, Brown S, MacNee W,
Donaldson K. (2008). Carbon nanotubes introduced into the abdominal cavity of mice show asbestos-
like pathogenicity in a pilot study. Nature Nanotechnology Online., 20 May, 2008.
Pooley FD, and Wagner JC. 1988. The significance of the selective retention of mineral dusts. Ann
Occup Hyg. 32(Suppl. 1): 187-194.
Rodelsperger K. (2004). Extrapolation of the carcinogenic potency of fibers from rats to humans. Inhal
Toxicol. 16:801-807.
3-52
-------
Stanton MF, and Wrench C. 1972. Mechanisms of mesothelioma induction with asbestos and fibrous
glass. JNatl Cancer Inst. 48:797-821.
Stayner LT, Kuempel E, Gilbert S, Hein M, Dement J. (2007). An epidemiologic study of the role of
chrysotile asbestos fiber dimensions in determining respiratory disease risk in exposed workers. Occup
Environ Med Online 20, Dec. 2007.
Wagner JC (1990). Biological effects of short fibers. Pp. 835-840 in: Proceedings of the Vllth
International Pneumoconiosis Conference, Part II. DHHS (NIOSH) Publication No. 90-108 Part II.
Dr. Gary Marsh
1- Do you agree that the data are sufficient to indicate that such differences may exist and an effort of
this type is warranted?
I would agree that the available experimental and human epidemiology data support the hypothesis that
the mineral type and size characteristics of asbestos are associated with markedly different risks for
malignant mesothelioma. While support for this hypothesis as it relates to the risk for lung cancer is
much less consistent, some recent work, such as the meta-analysis of Hodgson and Darnton (2000),
provides some compelling support for the hypothesis that lung cancer risks are related to asbestos fiber
type. Much of the evidence against this hypothesis for lung cancer stems from the epidemiology study
of the Charleston, South Carolina asbestos textile workers, and the meaning and significance of these
anomalous results remain a subject of scientific debate.
While the available epidemiology data provide support for this type of effort, it is apparent that we do
not have adequate environmental data within the epidemiology studies to enable scientifically sound
model fitting by fiber size bins. In particular, aside from the recent work of Stayner et al. (2007), we do
not have TEM-equivalent data for any of the existing epidemiology studies and the proposed PCM to
TEM conversion methods have been deemed by the SAB as too unreliable for practical purposes. This
effort should be re-visited as more TEM or TEM-equivalent data become available that are directly
related to human health outcomes.
Model fitting by asbestos fiber type (i.e., 2 bins) using the proposed methods appears to be feasible
based on the available epidemiological data. An alternative approach to this latter effort might be an
update of the meta-analysis performed by Hodgson and Darnton (2000).
2) Please comment on the adequacy of Section 4 (Overview of Human Studies) which serves as part of
the scientific bases for the proposed dose-response assessment approach.
As the authors admit, this section is relatively brief, providing a general overview of the published
studies that have reported adverse effects from human exposure to asbestos. The reader is referred to
several government agency reports (IARC, 1977), WHO (2000) and ATSDR (2001; 2004) for more
detailed reviews (it should be noted that many other excellent reviews are available in the peer-reviewed
literature). This brief summary succeeds in providing some basic background information on the main
non-cancer and cancer health outcomes as well as a brief overview of the role of fiber type with focus
on relatively more recent studies. The section on mesothelioma is particularly brief, especially
regarding studies of worker populations occupationally exposed to asbestos.
3-53
-------
While the brevity of Section 4 may be appropriate for the larger goals of the overall document, an
unnecessary and disconcerting disconnect exists between the literature for lung cancer and
mesothelioma summarized in Section 4 and the literature actually used in the modeling protocol, which
is described in general in Section 9 and in considerable detail in Appendix A. That is, many of the
"primary" and "other" literature cited in Appendix A and used in the modeling of these two health
outcomes were not cited in the general epidemiology overview in Section 4.
Because the epidemiological literature of human health effects from asbestos exposure is vast and can
defy attempts to summarize adequately, a logical and consistent approach is needed when selecting a
representative set of articles for background purposes. One simple alternative approach to the choice of
lung cancer and mesothelioma studies in Section 4 might be to include, as a minimum subset, all the
studies cited in Appendix A, as well as a representative selection of studies that did not meet the criteria
for inclusion in Section 9, but are nonetheless relevant. For example, as described in two recent review
articles (Goodman et al., 2004; Laden et al., 2004), a substantial number of epidemiological studies have
consistently demonstrated the absence of risk from lung cancer and mesothelioma among vehicle
mechanics exposed to low levels of chrysotile. The meta-analysis of Hodgson and Darnton (2000)
should also be discussed in detail in this document as it bears directly on the objectives of this risk
assessment.
Other areas that should be further developed in this section include: (1) the influence of fiber properties
(composition, length, wide, durability) on the risks of human lung cancer and mesothelioma; (2) an
overview of the animal toxicology data, especially the long-term inhalation studies, pertaining to
asbestos, other mineral fibers and synthetic vitreous fibers, and their implications regarding human
health risks; and (3) background information on the relevant asbestos exposures and perceived human
health risks related to Superfund sites (including an in-depth review of the background and completed
epidemiology studies at the Libby, Montana site) as this is the purported impetus behind this entire risk
assessment exercise.
3a Do you agree that the lung cancer and mesothelioma risk
models that are proposed are a scientifically valid basis for this model fitting?
This is a complex question, whose meaning was discussed extensively at the SAB meeting. My
response to what I believe this question is asking was essentially included in my response to question (1)
above. To reiterate, while the available epidemiology data provide support for this type of effort, it is
apparent that we do not have adequate environmental data within the epidemiology studies to enable
scientifically sound model fitting by fiber size bins. In particular, aside from the recent work of Stayner
et al. (2007), we do not have TEM-equivalent data for any of the existing epidemiology studies and the
proposed PCM to TEM conversion methods have been deemed by the SAB as too unreliable for
practical purposes. This effort should be re-visited as more TEM or TEM-equivalent data become
available that are directly related to human health outcomes.
Model fitting by asbestos fiber type (i.e., 2 bins) using the proposed methods appears to be feasible
based on the available epidemiological data. An alternative approach to this latter effort might be an
update of the meta-analysis performed by Hodgson and Darnton (2000).
3-54
-------
Section 9 - Epidemiological Data Proposed for Use
Section 9.1 - Criteria for Study Selection
11 a) Are the study specific selection rules proposed scientifically valid for the intended use?
The first and second proposed study selection rules (published in refereed journal and must provide data
that can be expressed in terms of the quantitative risk models for lung cancer and/or mesothelioma) are
straightforward, logical and scientifically valid. The wording of the second rule might be revised as
"The study must provide data that can be used directly or to provide accurate (reliable and valid)
estimates that can be expressed in terms of the quantitative ....", as this conforms better to the actual
procedures that are proposed.
The third rule (the study cohort must consist of individuals who were exposed to approximately the
same atmospheric composition of asbestos) is problematic as it is ill defined and does not seem to have
been applied consistently to all candidate studies to include/exclude subjects. Moreover, confusion
arises between the rule used to exclude studies due to exposure mixes and the detailed procedure
described in Section 10.3 to estimate the "fraction amphibole data" (as well as the detailed procedure
used to characterize the uncertainty in the fraction amphibole described in Appendix C). For example,
contrast the first statement on page 74 under "cohorts with mixed exposure",
". . . OSWER recommends that studies in which health statistics were combined across two or more sub-
cohorts exposed to substantially different workplace atmospheres should be excluded. "
and the statement on page 72, Section 9.1.3:
"Hence, studies in which the cohort is known to be composed of individuals who are exposed to
differing types of atmospheres are excluded from the data fitting process. "
with the procedure on page 79 under "chrysotile plus amphibole":
"In some studies, the description of the workplace and its operations makes clear that both chrysotile
and amphibole were used in the workplace. Ideally, data from TEM studies of air samples collected
from the workplace would serve as a basis for estimation of the relative amounts of chrysotile and
amphibole in the exposure atmosphere. However, in the absence of such data, information on the
relative amounts of the different types of asbestos purchased or processed can be used as a rough
surrogate for the relative amounts in the atmosphere. "
The authors need to clarify the distinction between excluding studies based on mixed atmospheres, and
including studies with mixed atmospheres and estimating the fraction amphibole. The clarification of
this distinction should include a detailed explanation of the reasons why each of the excluded candidate
studies was excluded.
3-55
-------
Should any additional selection rules be applied?
Another selection rule that might be considered by OSWER is that the studies meet some specified level
of data quality. As noted on page 69, this was recommended by some peer consultation panel members
who reviewed the approach of Aeolus (1999, 2001), but was rejected by OSWER as an impracticable
requirement. As noted on page 70, OSWER believes that the assignment of PDFs around each data
input item should account for any differing levels of data quality between groups and studies
lib) Is it appropriate to assume that all workers in a cohort are exposed to an atmosphere with a
constant composition (i.e., the mixture of asbestos types and sizes is constant) unless the authors report
information to the contrary? If this is not the appropriate assumption, what alternative strategy would
be available?
See the response to 1 la
lie) Should a set of minimum data quality requirements (other than those above) be established for
inclusion of a study in the analysis? If so, what elements of data quality should be considered and how
should those data quality rules be established?
One approach to this would be to require that included studies meet the minimum "Good Epidemiology
Practices or GEPs" described by Cook (1991). GEPs, which are now commonly applied in occupational
epidemiology projects, were modeled after their "Good Laboratory Practices or GLPs" counterparts in
the toxicology area, and provide standardized guidance regarding protocol development, reporting,
quality assurance of data maintenance and documentation of analytic procedures.
lid) For lung cancer, OSWER's approach requires that there be at least two exposure groups per study
in order to impose some constraint on the value of the study specific value of a. However, OSWER is
proposing to use data from three cohorts of described by Enter line and Henderson (1979) event though
there is only one dose group for each cohort. This is because a reliable estimate of afar the combined
cohort can be derived from the data of Enter line et al. (1987). Is this approach appropriate and
scientifically justifiable? If not, can you suggest an alternative strategy for retaining the data from this
important study or should this study excluded?
I believe this approach is appropriate and scientifically defensible although it may not be the best
approach. The error associated with this estimation should be adequately handled adequately by the
weakly informed prior for a described in Section 8.6.2. (i.e., each as is UNIFORM (0.1, 10)).
Another approach to estimating a study-specific as or global a would be to adjust the general population
rate for lung cancer by the likely positive confounding by smoking that would occur in a blue collar
working population. This could be done on a study-specific basis using the lung cancer rates of the
local population or more globally by using national lung cancer rates. Estimates of the positive
confounding by smoking could be obtained from available government reports or published
epidemiology studies and incorporated into the adjustment using Monte Carlo-based sensitivity analysis
(e.g., see Steenland and Greenland, 2004).
3-56
-------
lie) One key assumption in any meta-analysis is that the data sets included in the analysis are
homogeneous. How should the assumption of homogeneity be assessed prior to combining the data
from the studies or groups?
First, some of the better detailed discussions of basic meta-analysis methods and related tests of
homogeneity can be found in Bailey, 1987; Berlin et al., 1989; Deeks et al., 2001; DerSimonian and
Laird, 1986; Fleiss, 1993; Higgins et al., 2003 and Petitti, 2001. However, it remains unclear exactly
how these methods would apply or need to be modified for incorporation within the proposed risk
assessment methods.
With the above qualification in mind, briefly, a basic meta-analysis should include a formal test of
homogeneity of the risk estimates comprising a given meta-RR. Statistical heterogeneity of the RRs is
assessed by the I2 index, which describes the percentage of total variance across a given set of studies
that is greater than that expected by chance. The I2 index is calculated from the heterogeneity chi-square
(Q) and degrees of freedom (df) statistics from a meta-analysis (I2 = 100% x (Q-df)/Q), and is compared
to the chi-squared distribution with n-1 degrees of freedom, where n is the number of studies. Many
software packages are available to perform meta-analyses, including StatalO (Stata Corp., 2007).
If you recommend statistical testing, please provide guidance on the reliability of a decision based
solely on the test statistic.
With the above qualification in mind, because heterogeneity tests for meta-analysis of studies are
generally conservative (i.e., they have low power) and to avoid type II errors, it is often recommended
that a significance level of 0.10 instead of the more traditional 0.05. Also, some authors use these
qualitative terms to characterize values for I2: low (0-24%), moderate (25-49%), high (50-74%) and very
high (>75%) (Fayerweather, 2007).
If testing produces evidence of heterogeneity between some studies, what steps can be recommended?
With the above qualification in mind, when the null hypothesis of homogeneity is not rejected, then a
fixed effects model is preferred. If there is significant heterogeneity, a random effects model is
preferred. The fixed effects model assumes that any differences between study results are due solely to
chance, while in the random effects model the observed effect is assumed to vary around some average
with sampling error.
Random effect models should not be used to "explain away" heterogeneity. In the presence of
significant heterogeneity, efforts should be made to explore the reasons for the heterogeneity and/or to
reduce the level of heterogeneity. These efforts might include stratification by other study factors to
identify homogeneous subgroups or meta-regression, in which the characteristics of the studies or the
subjects of the studies are used as explanatory variables in a multivariate regression with the effect size
(or some measure of the deviation from the summary measure of effect) as the dependent variable.
3-57
-------
Section 9.2 and 9.3 - Studies Proposed for Use and Studies Excluded
12a) Are you aware of any studies that should be included in the model fitting effort that are currently
excluded or omitted? If so, what are these studies, and do they meet the requirements for study
inclusion?
The above recommendation that the model fitting be limited to fiber type bins notwithstanding, it seems
that the Stayner et al. (2007) paper that used TEM data to reevaluate risks in the South Carolina cohort
should be considered for inclusion in any model fitting effort that attempts to address fiber size.
Otherwise, I am not currently aware of other studies that should be included in the model fitting effort.
Per my response to question 1 la, it would be helpful to see a detailed discussion of the reasons why
each of the excluded candidate studies were excluded from the model fitting exercise. In other words, it
would be useful to see the entire list of candidate studies considered by OSWER for inclusion.
12b) Are there any studies that are currently proposed for inclusion in the analysis that you believe
should be excluded? If so, why?
As noted during the SAB meeting, a number of the proposed studies for evaluating lung cancer risks
utilized internal comparisons and thus do not provide an estimate of alpha. All the studies currently
included do appear to have satisfied the first two selection criteria, and as noted above, uncertainties
exist in the current document regarding the selection of studies based on mixed atmospheres.
12c) In cases where the epidemiological data are not reported in the form needed for use in the fitting
effort, are the methods used to estimate the exposures scientifically sound, and are the methods used for
characterizing the uncertainty in the estimates appropriate?
As noted above, while the available experimental animal and human epidemiology data provide
underlying support for this type of effort (i.e., support regarding differential risks for mesothelioma and
possibly lung cancer by fiber type and size), it is apparent that we do not have adequate environmental
data within the available epidemiology studies to enable scientifically sound model fitting by fiber size
bins. In particular, aside from the recent work of Stayner et al. (2007), we do not have TEM-equivalent
data for any of the existing epidemiology studies and the proposed PCM to TEM conversion methods
have been deemed by the SAB as too unreliable for practical purposes. This effort should be re-visited
as more TEM or TEM-equivalent data become available that are directly related to human health
outcomes.
One questionable area related to the availability of epidemiology data is how the model fitting process
for lung cancer will explicitly account for available (or unavailable) data on tobacco smoking histories.
Appendix A of the draft document describes the disposition of smoking history data for each of the
included studies. These data range from none to detailed data at the individual subject level. Without
any provision for the available data and considering the effect modification evidence for smoking and
asbestos, the model fitting will be combining higher risks from asbestos among smokers with lower
risks among non-smokers. It is also not clear why smoking data were not considered as a source of
uncertainty in Appendix C?
3-58
-------
Also, given the uncertainties surrounding the inclusion or exclusion of studies involving worker
exposures to mixed atmospheres (see response to question 1 la), it is not entirely clear how the modeling
of the potency factors will explicitly account for mixed atmospheres when they occur.
Dr. Luis Ortiz
Question 1) Do you agree that the data are sufficient to indicate that such differences may exist and that
an effort of this type is warranted?
The data available supports the notion that differences in asbestos type and particle dimension confer
significant differences in carcinogenic (as it pertains to lung cancer and mesothelioma) potency and
therefore, the proposed new approach appear to be warranted to more definitively resolve the issue.
However, this new proposed approach levels on the harmonization of pre-existent data (mostly
epidemiological reports from well characterized cohorts of asbestos exposed subjects) to adopt a multi
bin mathematical approach to estimate cancer risk according to mineral groups and measurements of
particle size based on transmission electron microscopy. Therefore, it appears that a number of
simplifying assumptions that were necessary during the harmonization process deserve further analysis
(even prospective validation) and as such the new approach should be considered interim.
Question 2) Please comment on the adequacy of Sections 2-5 of the overview of EPA 's OSWER Draft
Report.
Sections 2-5 are designed as a backbone to illustrate the needs of the OSWER report. These sections
should facilitate analysis of the OSWER document by the members of SAB and the broader public. A
potential consequences of endorsing such document is that the general public may equate this support
with that of and official change in EPA policy toward asbestos carcinogenic potency then it may be
advisable that such document be supplemented to cover a number of item as discussed below.
There must be ample discussion in the introduction of these sections stating that this is an interim
approach to assess the question of how physical difference in the composition of asbestos fibers modify
its cancer inducing capacity. A clear description of the historical and current EPA's needs that motivate
the OSWER report should be stated. Specifically, the document should make reference to the current
EPA priorities in the clean up efforts of superfund sites such as Libby, Montana and other sites around
the US (as discussed during the SAB meeting in Washington, D.C.).
Section 2 is adequate in length for the purpose of the OSWER report as it describes both the
fundamental aspects of the differences in asbestos fibers and clarifies basic aspects of the methodology
used to study the question at hand. However, a more detailed discussion of the importance of the
physical aspects of the asbestos fibers (length as well as width) and whether or not these properties bear
directly in the carcinogenic (lung or mesothelioma) capacity of the asbestos fibers is necessary.
Although Section 3 provides a summary of pertinent data linking the experimental exposure of animals
to asbestos with the subsequent development of cancer this section does not provide description of the
in-depth biology necessary to support the larger questions addressing the fitting of the published human
data as it pertains to the evaluation of whether differences in asbestos type and fiber dimension confer
significant differences in their ability to induce tumors (which is the question at hand).
This section of the OSWER report should also clearly describe that currently there is a definitive paucity
of scientific information, both in animals (Stanton and Wrench J. Natl. Cancer Inst. 48:797-821,1972) as
well as humans (Stayner LT et al. Occup Environ Med. 2007 Dec 20; [Epub ahead of print]) regarding
3-59
-------
the use of TEM to characterize fiber size specific asbestos exposure. In the case of Libby there are
samples that could be subjected to TEM analysis to properly address this deficiency. Therefore, this
aspect could be incorporated in the report as a scientific and investigational priority to the agency
Similarly, there appears to be room for improvement regarding the description of a body of published
work describing the factors that modify the environmental host interaction and determine individual
susceptibility to asbestos-induced cancer. Surprisingly, there is little mention of experimental animal
data addressing the relationship between smoking and its effects on the asbestos induced malignancy
(which is charged in question 3).
Sections 4-5 constitute a summary of studies describing the effects of asbestos in human beings and
provide an overview of the potential biological mechanism of the asbestos mode of action. Although
these sections are intended as an introductory summary to the reader of the OSWER draft report these
sections are superficial in relation to the importance of the work proposed by the OSWER and although
they summarize some of the key published data linking asbestos to both lung cancer and mesothelioma
they do not address in detail the issue of whether variability of fiber type confer differences in potency
towards carcinogenesis. The summary of the data linking asbestos to mesothelioma is particularly brief
and would benefit from adding additional references (some of which are subsequently included in
Appendixes). As was the case of the animal data, there are almost no descriptions of the potential
interactions of asbestos with smoking and certainly, no consideration to the relevance of individual
susceptibility as it pertains to the pathogenesis of lung cancer and mesothelioma.
Finally, the approach to the biology of the carcinogenic mechanisms of asbestos is timid, lacking in
molecular depth and not providing a biologic foundation to back the epidemiologic approach map (that
could be experimentally adopted to support the imminent fitting of the human epidemiological data) to
indicate how the proposed differences in the physical properties of the asbestos fibers determine their
carcinogenic potency. Therefore, although the number of references could be adequate for a report of
this magnitude, a more clear articulation of these references could be structured to reflect the historical
perspective that motivated the adoption of the fitting methodology proposed in the current report.
Question 3: Do you agree that the lung cancer and mesothelioma risk models that are proposed are a
scientifically valid basis for this fitting effort? Should additional model forms be investigated?
In general the approach is sound and so the modeling is scientifically valid. However, the development
of a progressive approach to determine the importance that asbestos fiber type and size confer to specific
risk estimates requires the availability of data sets derived with the use of transmission electron
microscopy (TEM) to appropriately describe the asbestos fiber size and length distribution to which
subjects are exposed. Currently, it appears that the only published data available is that of Stayner et al
(Occup Environ Med. 2007 Dec 20) characterizing the South Carolina textile cohort. Therefore, further
validation may be pertinent at this point and it appears that the superfund at Libby could offer a great
opportunity to validate the measurements of the amphibole particles by using TEM as appropriate filters
are available.
As presented in the OSWER document, these data are not available and therefore the working data sets
had been extrapolated from pre-existent cohorts (17 in total). These calculations assume a number of
simplifying assumptions that introduce bias. While the current OSWER report acknowledge most of
these bias and adopted stringent protocols to limit and correct them questions remain regarding the
ability of this fiber size specific TEM matrixes to account for within industry variation. Thus, as
3-60
-------
demonstrated in Dement et al 2007, application of TEM to a well described (previously characterized by
use of PCM-based fiber count) cohort of subjects from the Charleston, SC, textile plant identified
differences in asbestos fiber size among the different operations in this plant. Thus, the TEM-generated
matrix did not accurately predict inter occupational and industry specific variations of these
measurements.
TEM data sets appear to more adequately provide evidence that suits their use in assessing the risk of
asbestos-induced lung cancer. However, this does not appear to be the case with mesothelioma and
fewer TEM derived matrixes are available due to the prevailing limitations in inferring the equivalent
calculations.
Similarly, there appear to be inconsistencies in judging the rational for excluding the value of data sets
from specific cohorts to conduct risk assessment. A clear example of this, identified by the
correspondence shared by the members of the asbestos SAB, is the exclusion of the studies by Selikoff
and Seidman on insulator workers.
As stated in the OSWER report the current notion regarding the interaction between asbestos and lung
cancer is that this relationship is multiplicative. However, given the ongoing changes in the biology of
the lung cancer (increased number of subjects with lung cancer had not been smokers), the decrease in
the population exposed to smoking, then it appears that it is imperative that the nature of this
interrelation be revisited at this point. The proposed model appears adequate to do so and it appears
premature to look for alternative methods as no other strategy (ies) has been validated to assess the
asbestos conferred risk of lung cancer or mesothelioma models that are likely to be superior to the
current one.
Question 4: Is fitting at the group level (based on the number of cancer cases observed) preferred to
fitting at the study level (based on the study-specific KL or KM values) ?
The rational to support binning by fiber composition (amphiboles vs. chrysotile) and by fiber
dimensions (length or width intervals) is sound as the information available support the concept that
these variables may directly influence the biological effects of asbestos fibers inhaled into the lung or
deposited in the pleura. Otherwise, it was the general consensus at the SAB meeting that the OSWER
report did not provide a sufficient data to support binning based on fiber size. However, as stated above,
the SAB consider that the current efforts at Libby offer a unique opportunity to prospectively study this
important question.
Question 8: Do you agree that multiple binning strategies should be evaluated, or do you believe that a
physiological basis exists that can be used to identify a particular set of length and width cutoffs that
should be assessed? If so, what would these length and width cutoffs be, and can these bins be
implemented considering the limitations in the available TEM particle size data sets?
As discussed above the rational for this approach appears to be reasonable. However, the fundamental
issue that limits this approach is the lack of current available data to support the methodological
approach. As discussed in the SAB meeting currently, there is only one well-characterized cohort of
asbestos exposed subjects in which fiber characteristics have been analyzed by TEM (Stayner LT et al.
3-61
-------
Occup Environ Med. 2007 Dec 20) as such this limitation (paucity of validated data) decreases
enthusiasm for the approach.
I would also like to defer in this questions to the comments consigned at the SAB by Dr. M. Lippmann
in which he stressed the fact that while descriptions of bins according to length make sense for lung
cancer (Lippmann, M. 1994. Occup. Environ. Med, 51:793-8) they may not be suitable for studying the
causation of mesothelioma as fiber diameter determine the ability of asbestos fiber to traffic via
lymphatic channels to deposit into the pleura.
Questions 11 and 12: Are the study-specific selection rules proposed above scientifically valid for the
intended uses? Should any additional selection rules be added? Is it appropriate to assume that all
workers in a cohort are exposed to the same atmosphere with a constant composition (mixture of
asbestos types and sizes is constant) unless the authors report information to the contrary?
The reality is that there is a paucity of data to support such an enterprise and only one recently published
article has comprehensible TEM characterized asbestos fiber properties and correlated them to relevant
epidemiological data (Stayner LT et al. Occup Environ Med. 2007 Dec 20). Therefore, every effort
should be made to consider available information (transparent and scientifically valid data that could
undergo stringent and un-bias scrutiny) and incorporate this data if there is a valid scientific reason.
As mentioned above, recent information indicate that even data set obtained by using fiber size specific
TEM matrixes do not account for within industry variation (Dement et al 2007). Thus, this appears to
be an area ripe for research application to pertinent cohorts such as the Libby mining sites.
Questions 13 and 14: Is it scientifically justifiable to employ a default dust-to-PCM conversion factor
when there are no site-specific data available? Is it scientifically valid to use surrogate TEM data to
estimate bin-specific concentrations and exposure values in studies where these data are not reported?
If not, what alternative approach could be followed, or what additional data would be helpful?
The answer to this question is not because previous experience has shown that impinger data cannot
reliable used to generate PCM comparisons. Therefore, it appears that this limitation will also limit the
use of such data to estimate TEM fiber size distributions. Similarly, as discussed during the SAB
meeting, there are limitations to the ability of PCM estimates to produce consistent fiber size
distributions as measured by TEM.
Asbestos Charge Questions Section 3 (Toxicology) & Section 5 (mode of action) and
Please comment on the adequacy of these sections (3-5), which serve as the scientific bases for the
proposed dose-response assessment approach.
Sections 3-5 are designed as a backbone to illustrate the needs of the OSWER report. These sections
should facilitate analysis of the OSWER document by the members of SAB and the broader public. A
potential consequences of endorsing such document is that the general public may equate this support
with that of and official change in EPA policy toward asbestos carcinogenic potency then it may be
advisable that such document be supplemented to cover the following items.
1. There must be ample discussion in the introduction of these sections stating that this is an interim
approach to assess the question of how physical difference in the composition of asbestos fibers modify
3-62
-------
its cancer inducing capacity. A clear description of the historical and current EPA's needs that motivate
the OSWER report should be stated. Specifically, the document should make reference to the current
EPA priorities in the clean up efforts of superfund sites such as Libby, Montana and other sites around
the US (as discussed during the SAB meeting in Washington, D.C.).
2. A more detailed discussion of the importance of the physical aspects of the asbestos fibers (length as
well as width) and whether or not these properties bear directly in the carcinogenic (lung or
mesothelioma) capacity of the asbestos fibers is necessary.
3. The report should also clearly state that currently there is a definitive paucity of scientific
information, both in animals (Stanton and Wrench J. Natl. Cancer Inst. 48:797-821,1972) as well as
humans (Stayner LT et al. Occup Environ Med. 2007 Dec 20; [Epub ahead of print]) regarding the use
of TEM to characterize fiber size specific asbestos exposure. In the case of Libby there are samples that
could be subjected to TEM analysis to properly address this deficiency. Therefore, this aspect could be
incorporated in the report as a scientific and investigational priority to the agency.
4. Similarly, there appears to be room for improvement regarding the description of a body of published
work describing the factors that modify the environmental host interaction and determine individual
susceptibility to asbestos-induced cancer. Specifically, there is little description of experimental animal
data addressing the relationship (additive versus multiplicative) between smoking and asbestos exposure
on its carcinogenic effects.
5. Finally, the approach to the biology of the carcinogenic mechanisms of asbestos is timid, lacking in
molecular depth and not providing a biologic foundation to back the epidemiologic approach map (that
could be experimentally adopted to support the imminent fitting of the human epidemiological data) to
indicate how the proposed differences in the physical properties of the asbestos fibers determine their
carcinogenic potency.
Dr. Julian Peto
Leslie Stayner recognises that his response raises various crucial points in relation to fibre type and
other issues on which the Panel do not agree. I have therefore addressed his remarks directly in my
response (see below) rather than writing a separate commentary. Our response to the EPA should
identify crucial areas of disagreement. Whether amosite causes a greater lung cancer risk than
chrysotile is one such issue, although in the absence of adequate exposure data in fibre/ml longer than 5
microns in any of the epidemiological studies, let alone in subdivisions of length, this is not a well-
defined question.
This approach would clarify the points of disagreement and give the public an idea of who holds which
opinions, and why they hold them. My only plea is that we agree to distinguish clearly between
scientific evidence and consideration of the social or legal effects of the EPA's position. Both are proper
issues for discussion, but they must not be confused.
Two fundamental weaknesses in the proposed analysis were discussed by the Panel but are perhaps not
generally appreciated. The first is the assumption that the lung cancer risk is proportional to duration of
exposure. The draft EPA report states that this was established in the 1986 EPA report, but it would be
useful for the evidence on this to be reexamined systematically, particularly in cohorts exposed to
chrysotile. Amphibole exposures of less than 5 years duration have caused substantial lung cancer risks,
3-63
-------
but this does not seem to be true of chrysotile, perhaps because it disappears from the lung much more
rapidly.
The second and more important weakness is the virtual absence of any reliable measurements of the
historical exposure levels that caused substantial cancer risks in any of the cohorts. Fibre counts, let
alone counts in each range of fibre size, were non-existent before the 1960s, different methods of
measuring particles gave very different readings, and parallel measurements of particles and fibres gave
different ratios in different workplaces, and even in different areas within the same industry using the
same instruments. Last but not least, the most heavily exposed workers were not monitored at all.
Richard Doll and I discussed these problems at length in our UK report (Doll R and Peto J: Asbestos:
Effects on health of exposure to asbestos. Health and Safety Commission. Her Majesty's Stationery
Office, London 1985). There were no systematic historical data on most amphibole workers, and the
following extracts from our report illustrate the weakness of the best of the chrysotile studies.
S Carolina chrysotile textile factory: 'The exposure data obtained in this factory are less extensive than
appears at first sight. A total of 5576 samples were taken before 1975, but only 376 midget impinger
samples were taken before 1960, including 112 by a life insurance company and 81 by the US public
health service or state board of health between 1930 and 1945. It is difficult to know how representative
these were, and many activities, including fibre mixing with pitch forks in an area where there was no
dust suppression, were unmonitored.'
Quebec chrysotile mines and mills: 'The Quebec [exposure] data, which are the most extensive, began
to be collected in 1949 There were no routine measurements in the open pits, and few in the
underground mines, and the exposure estimates assigned to the substantial proportion of the cohort who
worked in these areas may be particularly unreliable. The conversion of high particle counts to fibre
counts is also difficult, as only 34 (5%) of the parallel samples [taken in the!970s]... exceeded 3
mppcf, while the estimated average exposure levels of men with 20 or more years service ranged from
4.2 mppcf for "low" exposure to 46.8 mppcf for "very high" exposure... .As in our own study [the
Rochdale textile factory] the lack of contemporary particle and fibre counts during the period when the
exposures that caused the highest observed excess risks occurred, together with the poor correlation
subsequently observed between particle and regulated fibre counts, make it impossible to quantify the
dose-specific effect at low fibre counts with much confidence.'
The basic data on which the proposed analyses depend are thus so unreliable even for chrysotile that we
concluded: 'We hesitate to suggest that the [exposure data] are insufficiently reliable to justify making
any quantitative extrapolation from past experience to the effects of current exposures. Nevertheless,
this may in fact be the case, and we may have to be satisfied with qualitative conclusions based on
knowledge of the direction in which progress has been made and epidemiological observations of the
effects of qualitatively different types of exposure.' In relation to the amphiboles we concluded: 'The
use of crocidolite and amosite (and, we assume, other types of amphibole asbestos) is, in practice, more
hazardous than the use of chrysotile, possibly because of the longer residence time of amphibole fibres
in the lungs, but possibly also for other reasons ... .No worthwhile data are, however, available to
enable quantitative comparisons to be made between the effects on humans of the different fibre types.
It can be concluded only that the use of amphiboles should be avoided whenever possible and that extra
precautions need to be taken when exposure to them occurs. It is possible, though we believe unlikely,
that the hazardous effects of chrysotile are mainly due to contamination with small amounts of
3-64
-------
tremolite.' I am not at all sure that the limitations in the exposure data that we drew attention to in 1985
have been overcome, but they should certainly be discussed more critically in a new EPA report than
they were in the last one.
I would like two further documents that are mentioned in my response to be attached to the
Committee's final report:
(1) My 1985 comments on the first draft of the 1986 EPA report (attached at the end of this document)
directly addresses the main points of disagreement.
(2) My 1981 report to the EPA on asbestos in schools (also attached at the end of this document) used
the same models for lung cancer and mesothelioma as the 1986 report and the current draft, and is also
directly relevant. It was never made public, and by 1985 it could not be traced in the EPA's archives.
Incidentally, I would like the fact that I developed these models to be acknowledged.
Response by Julian Peto to Leslie Stayner's comments on the EPA draft report:
Proposed Approach for Estimation of Bin Specific Cancer Potency Factors for Inhalation
Exposure to Asbestos
I agree with Leslie Stayner's comments on fibre dimension but I disagree with his statement that there is
little or no difference in lung cancer risk between chrysotile and the amphiboles. He offers no evidence
apart from a reference to the factory in S Carolina to explain why he disagrees with the majority of
independent scientists and the pooled data on this issue. The idea that even a few years of exposure to
chrysotile at levels averaging 1 fibre/ml or less could cause a substantial risk of either mesothelioma or
lung cancer flies in the face of all evidence. Amosite exposure can increase the lung cancer risk
substantially after brief exposure (Seidman et al 1979), but substantial chrysotile exposure has not been
shown to do so. As we mentioned in our 1985 report to the UK Health and Safety Commission (Doll
and Peto 1985), the studies that Richard Doll and I conducted in a Rochdale textile factory using mainly
chrysotile gave a dose-specific estimate for lung cancer of the same order as the S Carolina study (Peto
et al 1985). The SMR for lung cancer was about two in workers exposed for between 10 and 20 years at
levels that probably averaged between 5 and 10 fibres/ml, but in men who worked for 1-5 years the
SMR was not detectably increased (7 lung cancer deaths observed, 9.9 expected). Note that these
analyses were restricted to men employed after 1932, when conditions had greatly improved. There
were 13 lung cancer deaths compared with only 1.6 expected and 2 mesotheliomas in men with more
than 10 years of very heavy chrysotile exposure in this factory before 1932 (Peto et al 1985). There was
a large excess of respiratory cancer in the S Carolina factory, but the excess in men exposed for 1-5
years (McDonald et al 1983: 10 observed, 6.1 expected) was not statistically significant. In Quebec
chrysotile miners the lung cancer SMR was 2.4 in workers who were heavily exposed for more than 20
years in the 1930s and 1940s and 1.5 for 5-20 years of exposure (23 lung cancer deaths) but was not
detectably increased for less than 5 years (27 deaths) (McDonald et al 1980). Such data convinced most
independent scientists many years ago that amphiboles are more dangerous than chrysotile for lung
cancer as well as for mesothelioma, and the 1986 EPA report was wrong in its fixed view that there is
little difference in hazard. My formal comments to the EPA on that report (attached letter to Dennis
Kotchmar dated 13th Aug 1985) are directly relevant to the questions that we are now readdressing 23
years later. The report that EPA commissioned me to write on asbestos in schools in 1981 is also
attached. In it I applied the lung cancer and mesothelioma dose and time models that I had published in
1979, and these have been used in most subsequent Government reports in the US and Western Europe,
3-65
-------
including the 1986 EPA report and the current draft. Assuming that dose-response is linear, that the
mixture of fibre types in schools would be similar to that of US insulators' exposure, and that insulators
were on average exposed at 30 fibre/ml, I showed that the lifetime cancer risk at an average of 0.002
fibre/ml in schools would be of the order of 1 in 100,000. This report was not published, and a copy
could not be found in the EPA's archives when I mentioned this in 1985 while reviewing the draft 1986
report.
Leslie concludes by stating that any research by the EPA on differences in risk between asbestos fibres
that is published before a ban on the few remaining uses of chrysotile has been introduced is an insult to
those now dying of mesothelioma. They and their generation will in fact be completely unaffected by
any change in current exposure, and we cannot base our scientific conclusions on this dubious prediction
of their reaction. Some estimate of the number of future deaths that this ban might prevent or cause in
the US would be more relevant. That requires either more research or a consensus statement by panel
members on the plausible range of health effects of different policies on asbestos management based on
existing models and evidence. My personal view is that a ban is now irrelevant, and the important issue
is whether the risk from the amosite and crocidolite that remains in some ships and waste sites and in
many buildings can and should be abated.
Our report should be open about these scientific and political differences of opinion. We should include
an appendix comprising Leslie's statement, this one from me, and comments from other Panel members
on fibre type differences in hazard so that the public can judge the spectrum and credibility of expert
opinion for themselves. We cannot write a report that conceals these disagreements, and they need to be
clarified. The reservations about the weak epidemiological evidence on fibre size that we all expressed
at the Panel's public meeting on July 21-22 are already being misrepresented as doubts about
differences between chrysotile and amphiboles.
Dr. Christopher Portier
General Comments: Two key elements are not readily addressed in this proposal. First, what is the
scientific support for the underlying model. Because the entire analysis is based upon the assumed
model and models within this class, there should be some discussion of the scientific basis for this
model. This is confounded by the second issue, and that is the loss of data, especially for mesothelioma,
that results from the need to divide the exposure into the two separate forms. I view the epidemiological
data as falling into three separate datasets; everything appropriate for 1 bin analysis (group A),
everything appropriate for a 2 bin analysis (Group B) and everything appropriate for a 4 bin analysis
(Group C). I assume that group C is a subset of Group B which is a subset of group A. I would suggest
you do three analyses for the 1 bin case (Groups A, B and C), two analyses for the 2 bin case (groups B
and C) and a single analysis for the 4 bin case (group C). Comparing A to B to C for the 1 bin case tells
me what is the impact of study selection in this situation. Same for the 2 bin case. These types of
comparison will tell about the impact of study selection on the posterior distribution for the potencies. If
there is a substantial difference in central tendency or variance in the posteriors, the Agency should
consider not using the binning approach or develop a method that allows for all of the data to be used in
the analysis or in the validation of the analysis.
Charge Question 1:1 have insufficient detailed knowledge of all of the studies being evaluated,
especially on the clarity of the exposure, to provide any definitive opinion on this question.
3-66
-------
Charge Question 2: My answer to this question is no. I feel the authors could have spent a bit more
space on the toxicological data because the clarity of the differences in response for fiber size and type
could be better supported through a careful examination of these data.
Charge Question 3: There was considerable discussion of potential differences in absorption,
distribution and elimination of fibers as a function of size, chemistry and fiber type. As a long-term
goal, the Agency might wish to pursue the development of models for ADME as a tool for potentially
addressing differences in risk due to the different types of particles.
On the question of whether the models that are proposed are a scientifically valid basis for the fitting
effort, I believe it is. Given that all they have to work with is population-based epidemiological data,
the exercise is nothing more than a curve-fitting exercise to explore trends in the data. As such, these
models are as good any one might wish to use and better than some due to the long history of these
models in asbestos-related epidemiological studies. This is not to say the model is correct; there is no
doubt that this multiplicative model is a simplification of a much more complicated biological process.
The authors may wish to try other models to explore the implications of model choice on the eventual
predicted risks. Seeing no differences from a broad spectrum of models gives more confidence in the
results. Seeing large differences may lead to a discussion of potential assumptions that support one
model in favor of another or to additional data/experimentation that could lead to one specific answer.
Charge Question 4: The assumption of Poisson variation may be correct, but should be evaluated by
testing for over-dispersion. If there is over-dispersion, an appropriate change should be made to allow
for extra-Poisson variation.
Sections 8.4 - Characterizing Uncertainty In Exposure Data
In most cases, there are multiple sources of uncertainty in the measures of exposure reported in
published epidemiological studies. Section 8.4 provides an overview of how OSWER proposes to
characterize these uncertainties, and the details of the approach are provided in Appendix C.
Application of the proposed methods to each epidemiological study are presented in Appendix A.
Charge Questions 5a-5d:
5a. Have all of the important sources of uncertainty in cumulative exposure matrices been identified? If
not, what other sources should be accounted for?
As far as I can tell, yes. However, I am no expert on this.
5b. Is it appropriate to characterize the uncertainty from each source in terms of an independent
probability density estimated using professional judgment? If not, what alternative approach is
suggested?
Of course this assumption is wrong. The real question one has to ask is if this is appropriate in this case
or whether there is sufficient correlation between uncertainties that by treating these as independent one
is inflating the overall uncertainty. A sensitivity analysis of the distributions chosen for each factor will
tell you something about the relative importance of that factor. It would be wise to search for a data set
3-67
-------
that might tell you something about correlations between the most important factors and try using these
observed correlations to look at the sensitivity to correlation.
5c. Are the general strategies for selecting distributional forms and parameter values described in
Appendix C (and applied in Appendix A) appropriate for characterizing uncertainty in exposure
metrices? If not, what alternative strategies are recommended?
The approach seems appropriate. I am concerned that several distributions are based upon expert
judgment and ranges that seem to have no direct reference associated with them. These types of
judgments tend to be fairly optimistic and I suggest the sensitivity analysis look at this carefully.
In trying to find priors, my first choice would be to identify data that informs the prior directly, start
with an uninformed prior and use a hierarchical Bayesian model and update the prior in each iteration.
If this is not possible, use the scant data to develop an informed prior that is not updated. Failure to do
this, I would use an uninformed prior, do some sensitivity analysis and, for priors that seem to be
important, possibly se expert solicitation to identify a prior.
5d. Based on the assumption that each of the sources of error is independent, OSWER is proposing an
approach where the errors combine in a multiplicative fashion. Please comment on the scientific validity
of this approach and provide detailed suggestions for other approaches OSWER should consider.
There is really not much that can be said here about this approach. There does not seem to be a way to
test the assumption. Clearly other approaches (e.g. additive) could be used and would likely yield
different results. But these would also not appear to be testable with the available data.
Section 8.5. Fitting Approach
OSWER considered a wide range of strategies for fitting the epidemiological data to the risk models,
including simple minimization of squared errors, weighted regression, maximum likelihood methods,
measurement error models, Monte Carlo simulation, and Bayes-MCMC. Based on the recognition that
there is substantial error in both the independent variable (observed number of cases in an exposure
group) and the independent variable (metric of cumulative exposure for the group), OSWER is
proposing Bayes-MCMC as the most robust statistical approach for fitting the data.
Charge Questions 6a-6b:
6a. Is it appropriate to account for measurement error in the exposure data by using "measurement
error" models (weighted regression methods)? If so, how would the weights assigned to each exposure
value be assigned?
I am not certain why "weighted regression" enters into this? It is appropriate to use a measurement error
model to account for measurement error in exposure data from an epidemiological study.
6b. Is the assignment of a PDF for data quality sufficient or should data quality be factored into a
weighted likelihood analysis?
3-68
-------
Not sure what is being asked here since you have chosen the Bayesian approach. For the Bayesian
approach, the approach for including data quality should be fine.
6c. Do you think that the proposed strategy of fitting the risk models to the available epidemiological
data using Bayes-MCMC is scientifically justifiable? If not, what alternative strategy do you suggest,
and why?
The authors are very critical of the ML method but do not apply the same scrutiny to the Bayesian
methods. You basically can't get something from nothing. The Bayesian approach is giving you a
different answer and is based on a range of assumptions that may be very wrong but generally are not
testable. The rosie picture painted for the Bayesian approach should be more critically reviewed in this
document. That said, the Bayesian approach is appropriate.
Section 8.6.2 -Specification of Priors
Assuming that Bayes-MCMC is the method that will be used, it is necessary to specify prior uncertainty
distributions for each of the fitted parameters, including a (the vector of study-specific relative risks of
s
lung cancer at zero exposure), KL (the vector of bin-specific potency factors for lung cancer), and KM
b b
(the vector of bin-specific potency factors for mesothelioma).
Charge Question 7:
1. Are the priors prof
If not, what alternative priors should be considered, and why?
7. Are the priors proposed in Section 8.6.2 for a , KL , and KM consistent with available knowledge?
s b b
As far as this document goes, these priors seem appropriate. However, it seems to me that there might
be enough data to use an informed prior or even to set up a second hierarchy where the parameters for
the priors are informed by data. This would use more of the available data in a direct fashion.
Section 8.8 - Other Methods For Characterizing Goodness-of-Fit
OSWER is proposing that the initial evaluation of goodness-of-fit of different binning strategies be
based on the Bayes Factor, but is also proposing a number of additional evaluations to assess both
relative and absolute goodness-of-fit. These are described in Section 8.8.
Charge Questions 9a-9e:
9a. What method(s) is(are) preferred for characterizing the absolute goodness-of-fit of any selected
binning strategy? Should any of these methods be used to supplement the relative comparisons based on
the Bayes Factor? If so, how?
The proposed approaches seem appropriate.
9b. If different measures of goodness of fit do not yield results that agree, which method should be
preferred, and why?
That would depend entirely upon what was seen and why they differed. Question cannot be answered in
3-69
-------
the abstract.
9c. What methodological options do you recommend for validating the results of the modeling efforts?
What are the strengths and limitations of these options compared to others that might be available?
No comment.
9d. In lung cancer studies, it is expected that the value of a should be relatively close to 1.0. If the fitted
s
value of any particular value of a is substantially higher or lower than 1.0, should this be taken to reflect
s
that the data set giving rise to the value are somehow flawed or are too uncertain for use, and should be
excluded? If so, what criteria would you suggest for recognizing values that warrant concern?
You are placing too much focus on the nature of this statistic and not asking yourself what it really
means. If this value is substantially different than 1, it means you should return to your data set and
examine why this one data set is so different. Find what that difference is and THEN decide whether it
warrants exclusion from the overall analysis. Does it change your inclusion/exclusion criteria? Do you
need to reevaluate all of the cohorts?
9e. Is an examination performed of the residuals from the meta-analysis a rigorous and scientifically
valid assessment of homogeneity?
It is not rigorous, but it is useful.
Section 8.9 - Sensitivity Analysis
OSWER is proposing an approach for evaluating the sensitivity of the results to the various assumptions
and choices used in the effort that is based on series of "what if tests. For example, this may include
excluding all or some of the data from one or more of the studies, and assessing how those exclusions
impact the results. Likewise, one or more of the PDFs used to characterize uncertain input data may be
changed to evaluate if/how the results are altered.
Charge Questions lOa-lOb:
lOa. Is this "what if approach for evaluating sensitivity scientifically valid and useful?
Sensitivity analyses are indeed a valid and useful technigue for understanding the importance of
assumptions on the primary predictions from the modeling exercise. However, many times we approach
these types of exercise without any idea of what we plan to do with the results. The Agency needs to
give some thought to what will be done with the results of the sensitivity analysis.
The Agency has a fairly comprehensive approach to the sensitivity issue. The only additional analysis
we would suggest is that they consider varying the actual form of the model being applied to these data.
Some careful thought should go into this decision for alternative models prior to doing it, since alternate
models may demand alternate prior structures and the Agency would basically be conducting multiple
complete analyses of these data. The purpose of our suggestion is for the Agency to get a solid "feel"
for the impact of alternatives, not do multiple full analyses of these data.
3-70
-------
Dr. Carol Rice
Charge Question 2, section 4 (epidemiology)
Please comment on the adequacy of these sections which serve as the scientific bases for the proposed
dose-response assessment approach. [NOTE: is dose-response evaluated, or exposure-response?]
Non-cancer effects: section 4.1
It is not clear whether this section has been updated since the last version was presented to EPA.
One useful update is Rohs et al (2008). A PubMed search shows other citations which may be
considered for radiographic changes. Non-cancer mortality is also described in many of the cohort
studies, including an update of the Libby workers provided by Sullivan (2007).
Cancer effects: Section 4.2
The selection of the cohorts should be carefully evaluated in relation to the usefulness of the exposure
estimates in a "grouped analysis". While the methods in an individual study may provide useful internal
comparisons, the amount of estimation needed may lead to serious misclassification when studies are
combined, as is proposed. A careful review by industrial hygienists familiar with the sampling methods
used over time (especially impinger, MCE filter) and analyses of collected samples (particle counting,
fiber counting, fiber identification and sizing) is needed. Work provided to us by Stayner (see Dement
et al and references) provides a benchmark of procedures against which other exposure metrics can be
compared.
The Agency might consider seeking original data from studies other than the Charleston cohort, to
determine if the original metrics might be validated with additional data or other metrics might be
created. For example, exposures experienced by the Enterline cohort might be described by data
collected and stored by the corporation. A careful review by hygienists of original exposure data might
reveal other alternatives: comparison temporally in time, if not side-by-side measurements; examination
of differences between particle count by operation to replace a linear regression across all operations.
Should expert assessment be undertaken, the work of Dr. Gurumurthy Ramachandran would be very
important to consider (Ramachandran et al 2003; Hewett et al 2006).
Dr. Peto referred to a report describing further the Canadian exposure metric formulation; I have asked
him to forward that to me. An informative review of data available for many of the asbestos cohorts is
found in Gibbs (1994).
Role of fiber type: Section 4.3
From an industrial hygiene point of view, comparison of the exposure estimates is a challenge from
study to study. The proposal to use few conversion factors (including industry-wide factors for various
metrics) can be expected to introduce substantial error. It is likely that this will be non-differential
leading to an inability to find any real difference, but possible that an extreme example might result that
could lead to an erroneous conclusion (Wacholder, Dosemeci, Lubin,1991; Wacholder, 1995).
3-71
-------
The need for application of multiple conversion factors is illustrated in the recent Dement study, where
10 factors were used to convert from PCM-based estimates to estimates of exposure to fiber length and
width as would have been achieved if TEM analysis had been conducted on the original samples. A
recent report by Dodic-Fikfak (2007) of determination of conversion factors from mppcf (using a
Konimeter, not impinger) and filter samples for a single plant resulted in five different values,
depending upon operation.
The summary by Gibbs (1994), illustrates the need for multiple conversion factors. He describes the
4152 midget impinger samples from the mines and mills and 3096 mill dust samples used in the
Canadian studies. In later studies looking into conversion factors between impinger and filter samples
for the Canadian work locations, he noted that overall, the relation was only about 13% better than
random assignment; however, much of the variability was explained if job-site and mine-site was
considered. A review of papers cited by Gibbs may also be informative; see especially Ayer, Lynch and
Fanny (1965). The discussion in Enterline (1981) provides a perspective in terms of environmental
epidemiology.
Charge Questions 5,6,7
I was asked to provide some comment on the "inputs" to the model. See the sections above regarding
uncertainty in the exposure data, need for IH review and input into study selection and need for IH input
into the conversion factors.
Charge Questions 15 a, b.
15a. What method is best for estimating the uncertainty in lifetime cancer risk predications that are
associated with the uncertainty in the bin-specific potency factors?
Defer to Dr. Cox. The Agency might consider using only animal data as a "test of the model", as
exposure in many cases is more carefully characterized. This is of course not without uncertainty (e.g,
contamination, sizing bins) or experimental issues (e.g., dosing regime, time to sacrifice) or animal
sel ecti on/difference s.
15b. Assuming that estimates of exposure at Super fund sites will also have uncertainty, how should the
overall uncertainty in risk predications be characterized?
The level of uncertainty depends on the nature and extent of exposure information, and the context of
' estimates of exposure'.
At a generic Superfund site, there will be qualitative records of what has been placed in the site, core
samples, perimeter air samples and perhaps personal measurements of airborne particulate during
remediation. These provide little data in general to compare with a risk prediction. Prudent actions to
protect workers and public health will include limiting potential, unknown exposure.
3-72
-------
At Libby, there are well-characterized bulk and airborne material. These data are better than the input
into the risk prediction.
Citations for Charge Question 2
Epidemiology
Rohs et al. 2008. Low-level fiber-induced radiographic changes caused by Libby vermiculite, Am J
Respir Crit Care Med 177:630-637.
Sullivan PA. 2007. Vermiculite, respiratory disease and asbestos exposure in Libby, Montana: update of
a cohort mortality study.Environ Health Persp 115:579-585.
Wacholder S, Dosemeci M, Lubin JH. 1991.Blind assignment of exposure does not always prevent
differential misclassification. Am JEpid 134:433-437.
Wacholder S. 1995 When measurement errors correlate with truth: surprising effects of non-differential
misclassification. Epidemiology 6:157-61.
Baysian approaches to exposure assessment
Ramachandran G, Banerjee S, Vincent JH. 2003. Expert Judgment and Occupational
Hygiene Application to Aerosol Speciation in the Nickel Primary Production Industry. Ann Occup Hyg
47:461-475.
Hewett P, Logan P, Muhausen J, Ramachandran G, Banerjee S. 2006. Rating exposure control using
Bayesian decision analysis. J Occup Environ Hyg 3:568-81.
Conversion factors for asbestos exposure measurements
Ayer HE, Lynch JR, Fanny JH. 1965. A comparison of impinger and memberane filter techniques for
evaluating air samples in asbestos plants. Ann NY Acad Sci 132:274-287. (Note, incorrect volume
number cited in Gibbs—132 is correct).
Dement JM, Kuempel ED, Zumwalde RD, Smith RJ, Stayner LT, Loomis D. 2007. Development of a
Fiber Size-Specific Job-Exposure Matrix for Airborne Asbestos Fibers. OEM. 11/5/2007 (online).
Dodic-Fikfak M. 2007. An experiment to develop conversion factors to standardise measurements to
airborne dust. Arh Hig Rada Toksikol 58:179-185.
Enterline PE. 1981. Extrapolation from Occupational Studies: a Substitute for Environmental
Epidemiology. Env Health Persp 42:39-44.
3-73
-------
Gibbs GW. 1994. The Assessment of Exposure in Terms of Fibres. Ann occup Hyg 38:477-487.
Dr. Leslie Stayner
Following are my specific comments on the charge questions that I was asked to respond to (1, 2, 3, 4,
11 and 12). As requested, I have also included at the end of my comments several recommendations
for future research. I also want to make some general overarching comments, since I am afraid that my
main concerns might be lost in the details of my responses to these questions.
First of all, I unfortunately do not believe that there is adequate epidemiologic data to support the
development of the bin specific risk assessment models that EPA is seeking to develop at this time. I
believe that most if not all of the participants in our meeting agreed on this point. The use of data on
fiber size dimensions from one study to estimate fiber size dimensions for another is simply not
credible. At this point there is only one epidemiologic study that provides fiber size specific results,
which is our reanalysis of the South Carolina cohort [Stayner et al. 2007]. Although this study provides
support for the concern that long and thin fibers may be more strongly associated with lung cancer
mortality than short and thick fibers, the correlations in the exposures to different fiber sizes make it
impossible to determine their relative potency. There is one more study underway by Dr. John Dement
at Duke University of North Carolina textile workers that will be using essentially the same
methodology as our study to conduct fiber size specific analyses. It is possible that in the future we
may combine the data from these 2 studies and be able to obtain more definitive estimates of the relative
potency of different fiber sizes. However, this may still not be a sufficient amount of data and both
studies are of primarily chrysotile exposed workers and thus will not provide information on
amphiboles. This is why I strongly recommend that EPA support efforts to study other populations
where it would be possible to characterize the fiber size distributions, and in particular of workers at the
Libby facility.
Secondly, I do not believe that a good case can made that the potency of asbestos fibers for lung cancer
varies by fiber type. The only analysis that supports this conclusion is the Hodson and Darnton paper.
However, in reaching this conclusion they disregarded the findings from South Carolina cohort as an
outlier even though most observers would rate this study as having one of the best exposure assessments
of any of the studies. The fundamental and yet unresolved issue is that there is a large degree of
disagreement in the results from the studies (i.e., heterogeneity), and particularly between the findings
from the South Carolina and Quebec cohorts. I believe that the disagreement in the results between the
two studies may in part be explained by differences in the fiber size distributions. In fact in our study
we did find a higher percentage of long and thin fibers in South Carolina [Stayner et al. 2007] than what
had been reported in previous studies of Quebec mines and mills [Gibbs and Hwang 1975, Gibbs and
Hwang 1980]. If possible, further analyses of samples from the Quebec mines and mills would be
extremely helpful in attempting to resolve this issue. It is also quite possible that the discrepancy in
findings between these two studies may be explained by errors in the exposure estimates for the Quebec
study, which could have weakened the observed exposure-response relationship. The lack of correlation
between mppcf and fiber/cc measurements in the Quebec studies does suggest that there would have
been substantial misclassification of exposures from the conversion to fiber counts. In any case, I do not
believe at this time that is appropriate for EPA to make an assumption in its modeling that there is a
difference in lung cancer potency for different fiber types.
3-74
-------
Finally, I want to make a comment and suggestion to EPA regarding the extremely hostile reactions we
heard from many of the public commentators at our meeting regarding this proposal. I believe that in
moving forward with this proposal while not supporting a ban on asbestos this administration is in a
sense adding insult to injury to the workers and communities who have been victims of past exposures
to asbestos. The primary effort of the EPA at this time should be to convince this administration and
members of our Congress to pass proposed legislation banning the use and production of asbestos
products in our country. This ban would not only insure that we will never repeat our tragic mistakes of
the past, but will also send a strong signal to other countries in the world who are currently importing
and using asbestos in large quantities. Of course, I recognize that a ban will not address the issues of
natural occurring asbestos or the remnants of asbestos in hazardous waste and buildings in our country
and that a more reliable risk assessment model would be extremely useful for prioritizing the cleanup of
these situations. However, until such an asbestos ban is passed it is difficult to see how modifications in
the risk assessment models for asbestos will not be viewed with deep suspicion by the victims of
asbestos in this country.
Charge Question 1
Do you agree that the data are sufficient to indicate that such differences may exist and that an effort of
this type is warranted?
This question needs to be divided into two questions. The first part being is there sufficient information
to suggest that there are differences in potency of asbestos by fiber type and dimensions. The second
part of this question is whether such an effort as described in this proposal is warranted. I will provide
my views on each of these questions separately below.
Is there sufficient information to suggest that there are differences in potency of asbestos by fiber type
and dimensions?
I agree that there is evidence that asbestos fiber potency varies by mineral type and particle dimensions.
However, my confidence in this statement varies by cancer type. There is I believe a general consensus
in the scientific community that chrysotile asbestos is less potent than amphiboles for mesothelioma.
Support for this was discussed in a review we published in 1997 [Stayner et al. 1997]. This view was
strongly supported by the reanalysis of the Davis rat inhalation studies based on TEM data by Berman et
al. [1995], by the meta-analyses of the epidemiologic literature that was conducted by Hodson and
Darnton [2000], and by Berman and Crump [Aeolus] in their 2003 report developed for the EPA. All
of these analyses revealed highly statistically significant evidence that chrysotile is less potent than
amphiboles for mesothelioma. It also worth noting that the analysis by Hodson and Darnton also
indicated that amosite is less potent than crocidolite for mesothelioma.
The role of fiber size in mesothelioma induction is less clear. As recently reviewed by Dodson [2003],
the results from toxicological studies have been conflicting with some studies indicating long fibers, and
other studies indicating short fibers may be play a role in mesothelioma induction in animals. Short thin
fibers are the predominant type found in pleural tissues in studies of workers by Suzuki [2001].
3-75
-------
However, I don't believe that what is found in the either pleural or lung tissues can necessarily be
inferred to be the fibers that actually cause the disease (either for mesothelioma or lung cancer).
For lung cancer, I do not believe the available evidence convincingly demonstrates that mineral type is
an important determinant of potency. My basis for this conclusion was discussed in the 1997 review
article [Stayner et al. 1997]. Nothing that has been published since our review article has changed my
views on this subject and if anything I believe there is now further support for this position. The
analysis of the Davis rodent inhalation studies using a TEM analysis by Berman et al. [1995] failed to
demonstrate evidence of a difference in lung tumor potency by fiber type. In their meta-analysis of the
epidemiologic studies that they performed for EPA in 2003 [Aeolus], Berman and Crump also failed to
find significant evidence that lung cancer mortality varied by fiber type. Although their analysis did
suggest that potency was slightly less than amphiboles (26 to 42% in 3 different models) these
differences are likely to be explained by random error given the fact that these differences were not
found to be even close to statistically significant (p=0.23 to 0.51 in 3 models). In addition, these
differences are so small that they are not important from a public health viewpoint.
A much larger difference in lung cancer potency was reported in the metanalysis reported by Hodson
and Darnton [2000]. They suggested that chrysotile asbestos was 10 to 50 times less potent than
amphiboles. However, as they suggest the interpretation of their findings is greatly complicated by the
large difference in lung cancer potency derived from the studies of South Carolina textile workers, and
Quebec miners and millers. There was strong evidence of heterogeneity in the slope for lung cancer in
the Hodson and Darnton metanalysis that was largely attributable to the differences between the Quebec
and South Carolina studies. Hodson and Darnton chose to in effect ignore the results from the South
Carolina study in developing their estimates of lung cancer potency for chrysotile asbestos. They
reasoned that "Taking account of the excess risk recorded by cohorts with mixed fibre exposures
(generally, 1%), the Carolina experience looks untypically high." However, I think it is a very
questionable judgment to ignore the findings from the South Carolina study when in fact, as was noted
in the Berman and Crump report for EPA in 2003, this was one of the studies with the best information
on asbestos exposure levels. Actually I would think a stronger justification could be made for deleting
the Quebec study because of the large degree of uncertainty that exists in their exposure estimation
procedures related to their conversion from particle count measurements to fiber counts as discussed at
our meeting. I have previously demonstrated in an analysis that I presented at the EPA 2003 meeting in
San Francisco that if one deletes the Quebec study the result of the meta analysis would indicate that the
potency for chrysotile appears to be greater for chrysotile than amphiboles. I would not advocate
deleting the Quebec study, but this merely suggests that the results from meta-analyses are not robust
and dependent on whether one includes the Quebec or the South Carolina studies. Drawing conclusions
on relative potency for lung cancer will not be possible in meta analyses of the epidemiologic data until
we have explained the large differences in lung cancer potency between these two studies.
In contrast to fiber mineralogy, I do believe that there is substantial evidence to indicate that lung cancer
potency is likely to vary by fiber dimensions. Stanton [1981] was the first to suggest that long and thin
fibers might be more potent than thick and short fibers in inducing lung cancer based on pleural
implementation studies in rats. Inhalation studies of chrysotile asbestos exposure in rats conducted by
Davis et al. [1988] confirmed that long fibers were more potent in inducing lung tumors than short
fibers. The reanalysis of the Davis studies using TEM analysis by Berman and Crump [1995] provided
strong evidence that long and thin fibers were most strongly related to lung cancer risk. These
toxicologic findings are supported by the results from our recently published study of the South Carolina
3-76
-------
cohort using TEM based size specific estimates of chrysotile exposure [Dement et al. 2007 and Stayner
et al. 2007]. In this study we reported that cumulative exposure to long and thin fibers were most
strongly associated with lung cancer mortality. It should be noted, however, that short and thick fibers
were also associated with lung cancer mortality and that the interpretation of the study findings were
complicated by the strong correlations between the different size specific exposure measures. Although
I agree that the potency of asbestos fibers is likely to vary by fiber dimensions, I do not believe that we
have adequate data at this time to quantify these differences.
Is the effort as described in this proposal warranted?
Although I am sympathetic to the need for modification to our current risk assessment paradigms for
asbestos, I do not believe that the approach outlined in this proposal would yield valid or useful results.
The simple reason for my skepticism in this regard is that the data currently available to modify these
models is grossly inadequate for this purpose. The primary problem lies with the data available to
estimate the TEM specific levels of exposure for the cohort studies included in this analysis. Using
TEM data from one study setting to estimate fiber size distributions at another facility and even worse
for another industry is simply not reliable. This was clearly the opinion of the members of our panel
with expertise on TEM analysis who answered no to question 4 b in our charge (Is it scientifically valid
to use surrogate TEM data to estimate bin specific concentrations and exposure values in studies where
data are not reported?). The 2003 report prepared for EPA by Berman and Crump [Aeolus] also
provides empiric evidence that this approach doe not work since the addition of fiber size to their
models resulted in at best a marginally significant improvement in model fit for lung cancer [p=0.04 to
0.1] depending on model], and mesothelioma [p=0.05 to 0.24].
As far as estimating difference in potency by fiber type for lung cancer the EPA approach will face the
same problems as encountered in the previous meta-analyses with regard to the large extent of
heterogeneity in the study findings and the unexplained differences in potency for lung cancer risk
observed in the South Carolina and Quebec studies.
It is important to recognize that although the report includes an extensive effort to estimate the
uncertainties underlying the estimation of bin specific and fiber type exposures, that this effort can not
correct for these errors and could even conceivably introduce biases into the estimation of the model
parameters. There is so much uncertainly about how to estimate the pdfs for their uncertainty analysis
that I am concerned that their misspecification of the pdfs could actually introduce error into the
estimation process.
Given these concerns, I think it would be irresponsible for EPA to move forward with this risk
assessment model until we have more reliable data on which to base such an assessment. I do not
accept the argument that if EPA conducts this analysis it would be viewed by the agency and perhaps
more importantly by the public as "hypothesis generating exercise". Coming from a government
agency, the results from this assessment will undoubtedly be used by lawyers and policy makers. The
testimonies of several lawyers at our meeting demonstrated how even the 2003 Berman and Crump
analysis is being currently used in legal cases. Furthermore, it is not even credible that results from the
analysis would not be used by EPA since it is stated in the overview of the report (page 1) that OSWER
is proposing to use this as "an interim approach to account for the potential differences of cancer
potency between different mineral types and particle size distributions."
3-77
-------
Charge Question 2- Sections 6 & 7:
Sections 6 and 7 provide an accurate summary of the risk assessment methods that were used in the EPA
1986 and the Berman and Crump analysis that was conducted for EPA. Berman and Crump have
developed manuscripts for three papers based on their work for EPA, which I believe they are
submitting for publication in a peer review journal. There may be some slight modifications in these
papers that should have been noted in this document.
As a Co-Chair of the 2003 panel that reviewed the EPA proposal in 2003,1 do not feel the description of
our conclusions is accurately portrayed in this report. This report suggests (page 46) that the
"consultation panel generally endorsed the basic idea of a multi-bin approach". Although I think this is
true conceptually, several members of the panel had very serious misgivings about our ability to develop
such a model based on the data available at that time. This report then lists a number of issues and
limitations of the 2003 report, but it seems to attribute these comments to our panel. Our panel made
individual comments and I don't believe this is a fair summary of all of our concerns. One over riding
concern that I had at the time was that there was insufficient epidemiologic data to support the
development of a fiber size specific model. This unfortunately remains my chief concern today. I also
expressed a concern that the 2003 model provided fiber type specific slopes for lung cancer risk even
though their own analysis did not support their being significant evidence for this.
Charge Questions 3a-c
3a. Do you agree that the lung cancer and mesothelioma risk models that are proposed are a
scientifically valid basis for this fitting effort?
I believe the EPA 1986 models are a reasonable starting point for this effort. Crump and Berman
[Aeolus 2003] in their 2003 report for the EPA tested the the EPA 1986 models using raw data from the
South Carolina, Quebec and Wittenoon study cohorts and found that these models provided reasonably
good predictions for these studies.
3b. Should additional model forms be investigated?
While it would be nice to evaluate biologically based models, such as the 2 stage clonal expansion
model, as was suggested by Dr. Cox, I do not believe that this is possible given the data that is available
to EPA. It would also be desirable to test the assumption that cumulative exposure (duration x
intensity) is an appropriate metric for the lung cancer model as was suggested by Dr. Peto. Again I
don't think the data exists in the papers to allow separate modeling of duration and intensity.
3c. For lung cancer, the current risk model is multiplicative with the risk from smoking and other causes
of lung cancer. Should the nature of the interaction between asbestos and smoking be further
evaluated?
Yes I believe this relationship should be further evaluated. The observation that there is a multiplicative
effect of smoking and asbestos is largely based on the results from the study by Selikoff of insulator
workers [Selikoff et al. 1968]. There in fact is a tremendous variation in the evidence regarding whether
this relationship is multiplicative or additive [see Steenland and Thun 1986 and Vainio and Boffetta
3-78
-------
1994]. A recent analyses of this issue suggests that in fact the relationship may be somewhere between
additive and multiplicative [Wraith and Mengersen 2008].
4a. Is fitting at the group level preferred to fitting at the study level?
I do not believe that fitting at the group level should be preferred to fitting at the study level. The
primary advantage of fitting at the group level appear to be that it permits a better description of the
uncertainties involved. However, this advantage does not outweigh the serious loss of information that
results from grouping of the data. For several of the studies (South Carolina, Quebec and Wittenoon), it
is possible to obtain the raw data and to directly fit a slope (KL or KM) for these studies. Using group
categories for these studies results in substantial misclassification of exposures since it is necessary to
assign a midpoint or mean for each of the categories. In act Berman and Crump in 2003 were able to
obtain the raw data from these studies, and I would urge the EPA to use these individual data in
subsequent analyses.
4b. If so, is it scientifically justifiable to use a Poisson likelihood model for the observed number of
cases in each group?
Yes it is scientifically justifiable and appropriate to use a Poisson likelihood for the analysis of grouped
data. If the individual study results are modeled instead, as I would suggest, one could probably
assume a normal or more likely a log normal distribution applies. In any case, it will be important to
test the adequacy of the distributional assumption made.
Charge Questions 11 & 12
11 a. Are the study-specific selection rules proposed above scientifically valid for the intended uses?
Should any additional selection rules be added?
I do not believe these selection rules are appropriate. The first rule that a study has to be published in a
peer reviewed journal is in principle okay, but is not okay as it was applied. This rule was used as a
basis for rejecting the use of the raw data from South Carolina, Quebec, and Wittenoon which was used
in the previous analysis conducted by Berman and Crump for EPA in 2003. Although the findings from
these studies have all been published in peer review journals, the report rejected using the data from
these studies because the new analysis is not published. I believe this is being too restrictive and results
in the loss of some valuable information for the purposes of this risk assessment.
The second criterion requires studies to provide information that can be expressed in terms of the
models selected is reasonable in principle, but it also too restrictive as applied. Case-control studies and
other studies that could not be used to estimate the alpha were rejected on this basis. However, the
alpha is a correction parameter and one that is not necessary in studies where there is not likely to be a
difference in the background rates of diseases in the exposed and unexposed groups. For most case-
control studies alpha would be expected to be one if the study is properly designed.
The third criterion is unclearly written, but my understanding from the meeting was that this criterion
was primarily intended to exclude studies where the percentage of different fiber types or dimensions
was not constant over time. This was the basis for the exclusion of the Selikoff study of insulators.
3-79
-------
This study was used in the 2003 report by Berman and Crump. It seems it would be preferable to
include it, and express this issue is an uncertainty in the analysis.
As a general principle, I believe it is better not to exclude too many studies in a meta-analysis. Rather I
think the approach should be to investigate whether weaknesses and strengths of the study designs are
explanations of the heterogeneity observed in the analysis.
lib. Is it appropriate to assume that all workers in a cohort are exposed to the an atmosphere with a
constant composition (i.e., the mixture of asbestos types and sizes is constant) unless the authors report
information to the contrary? If this is not an appropriate assumption, what alternative strategy would be
available?
No, I do not believe it is appropriate to assume the composition (fiber size and type) is a constant in the
study if the authors do not state otherwise. Fiber dimensions were not found to be constant over
different operations in our study of South Carolina chrysotile textile workers [Dement 2008]. In many
study situations the fiber size dimensions would be likely to vary over time. The use of fiber types also
would be expected to have varied substantially in many industries over time. The only alternative
would be to limit the analysis to studies with well documented information on how the fiber size and
type varies over time and by operations in the study facilities. Unfortunately this only available for one
study, which is the recent study of South Carolina textile workers [Stayner et al. 2007] or from the
toxicologic studies?
Research Recommendations
1. Meta and Pooled Analyses
The Agency should consider conducting or funding a new meta-analysis to further evaluate the evidence
that asbestos cancer potency varies by fiber type. This effort is important because there are new studies
that were not incorporated in the most recent meta-analyses conducted by Hodson and Darnton or
Berman and Crump [2003]. It is critically important that these analyses fully assess heterogeneity and
possible explanations for any heterogeneity.
The Agency should also consider the possibility of funding a pooled analysis in which the raw data from
the individual epidemiologic studies are combined. Such an effort would be a substantial improvement
over a meta-analysis since the analysis would not be restricted to the categories of exposure and other
covariates that were reported in the different studies. It would also be possible in this analysis to
evaluate the assumption that cumulative exposure (product of duration and intensity of exposure) is an
appropriate metric for the development of the lung cancer model. A good model of how this study
might be conducted is the IARC pooled analysis for exposure to silica [Steenland et al.
2001]. In fact the IARC is uniquely well positioned to conduct such international collaborative efforts,
and I would highly recommend pursuing this possibility with them. I have already had a preliminary
discussion with one scientist at IARC (Dr. Kurt Straif) and he was very interested in this possibility
particularly since IARC is re-evaluating asbestos next year.
2. Libby Montana
There is a critical need for analyses of more epidemiologic studies using TEM based fiber size specific
estimates of exposures as was conducted in the recently published South Carolina textile cohort study
3-80
-------
[Stayner et al. 2007]. Based on discussions at our meeting it appears that there are old sampling filters
and slides that may be reanalyzed using TEM methods to develop a fiber size specific exposure matrix
for the study of workers that was recently updated by Sullivan [2007]. It may also be possible to do
fiber size specific analyses with the community but I think this is unlikely given the limited number of
cases of respiratory disease that have been reported. It may also be possible to find other asbestos
cohorts where a fiber size specific analysis could be conducted.
3. Medical Screening & Treatment
There is to my knowledge hardly any current research on treatment for asbestos related diseases. I
believe this is a very important and promising area for future research. Mesothelioma incidence in the
U.S. has just begun to peak and thus there are still many workers and community exposed individuals
who will develop this disease in the future. There also remains a large burden from lung cancer and
asbestosis in our country. Research on secondary prevention of these diseases either through early
detection or treatment at an early preclinical stage of the disease has great potential for preventing the
further development of asbestos related diseases.
References
Bermann DW, Crump KS, Chatfield EJ, Davis JM, Jones AD. The sizes, shapes, and mineralogy of
asbestos structures that induce lung tumors or mesothelioma in AF/HAN rats following inhalation. Risk
Anal. 1995 Apr (15(2):181-95.
Berman DW and Crump KS. New metrics for assessing asbestos-related cancer risk that address fiber
size and mineral type. Unpublished report. 2003.
Davis JM, Jones AD. Comparisons of the pathogenicity of long and short
7 fibres of chrysotile asbestos in rats. Br J Exp Pathol 1988;69(5):717-37.
Dement JM, Kuempel E, Zumwalde R, Smith R, Stayner L and Loomis D. Development of a fiber size
specific job-exposure matrix for airborne asbestos fibers. Occup. Environ. Med. published online 5 Nov
2007.
Dodson Asbestos fiber length as related to potential pathogenicity: a critical review. AJIM 44:291-297
(2003).
Gibbs GW and Hwang CY. Physical parameters of airborne asbestos fibres
in various work environments-Preliminary findings. Am Ind Hyg Assoc J 1975;36(6):459.
Gibbs GW and Hwang CY. Dimensions of airborne asbestos fibres. In:
Biological Effects of Mineral Fibers. JC Wagner, Ed. IARC Scientific
Publications, Lyon, France 1980.
3-81
-------
Selikoff I, Hammond E, Churg J. Asbestos exposure, smoking and neoplasia. Journal of the American
Medical Association 1968; 204:104-110.
Stanton MF, Layard M, Tegeris E, Miller E, May M, Morgan E, Smith A. 1981.
3 Relation of particle dimension to carcinogenicity in amphibole asbestoses and
4 other fibrous minerals. J Natl Cancer Inst 1981;67:965.
Stayner LT, Dankovic D, and Lemen R. Response to letters from Wagner JC on Asbestos Related
Cancer and the Amphibole Hypothesis. Am J Pub Health, 1997:87(4);687-8.
Stayner LT, Kuempel E, Gilbert S, Hein M, Dement J. An epidemiologic study of the role of chrysotile
asbestos fiber dimensions in determining respiratory disease risk in exposed workers. Occup Environ
Med. 2007 Dec 20; [Epub ahead of print]
Steenland K, Mannetje A, Boffetta P, Stayner L et al. Pooled exposure-response analyses and risk
assessment for lung cancer in 10 cohorts of silica-exposed workers: an IARC multicentre study. Cancer
Causes and Controls, 2001;12;773-784.
Steenland K, Thun M. Interaction between tobacco smoking and occupational exposures in the
causation of lung cancer. Journal of Occupational Medicine 1986; 28:110-118.
Sullivan PA. Vermiculite, respiratory disease, and asbestos exposure in Libby, Montana: update of a
cohort mortality study. Environ Health Perspect. 2007 Apr; 115(4):579-85. Epub 2007 Jan 3. .
Suzuki Y, Yuen SR. 2001. Asbestos tissue burden study on human malignant mesothelioma. Ind Health
39:150-160.
Vainio H, Boffetta P. Mechanisms of the combined effect of asbestos and smoking in the etiology of
lung cancer. Scand J Work Environ Health. 1994 Aug;20(4):235-42.
Wraith D, Mengersen K. Assessing the combined effect of asbestos exposure and smoking on lung
cancer: aBayesian approach. Stat Med. 2007 Feb 28;26(5): 1150-69.
Dr. James Webber
I have split my comments into two sections. The first section, A. Charge Questions, is devoted to
answering specific charge questions. The second section, B. Need for Future Research, is on the final
page and sketches my thoughts on what must be done before the EPA can consider going forward with
their OSWER risk-assessment proposal.
A. Charge Questions
Below, I have answered the charge questions on which I serve as primary and secondary lead. In
addition, I have responded to other charge questions where I felt I could make meaningful contributions.
Charge Question 1: Do you agree that the data are sufficient to indicate that such differences may exist
and that an effort of this type is warranted?
3-82
-------
No. A careful review of existing data reveals that environmental exposure measurements are currently
insufficient and/or inadequate for developing a new risk-assessment model. Conversions of impinger
(dust) concentrations and even PCM (fiber) data into (TEM-equivalent) fiber-size distributions cannot
be considered reliable because of the orders-of-magnitude uncertainty at each conversion step. At
present, only the data of Stayner et al. (2007) provide dependable fiber-size distributions that are
associated with known lung cancer and mesothelioma outcomes. As more TEM-based environmental
exposure data that are directly associated with health outcomes become available, this effort could be re-
visited.
References
Leslie T Stayner, Eileen Kuempel, Steve Gilbert, Misty Hein and John Dement. An epidemiologic
study of the role of chrysotile asbestos fiber dimensions in determining respiratory disease risk in
exposed workers. Occup. Environ. Med. published online 20 Dec 2007;
Charge Question 2: Please comment on the adequacy of these sections which serve as the scientific
bases for the proposed dose-response assessment approach.
Section 6
The conversion factor in Section 6.2 ("k", page 34) for "relative mesothelioma hazard" includes
RMHaii (page 35). RMHan excludes friction-product data because of their "uncertainty in the
values", where KLPCM is lung cancer potency. This exclusion is unwarranted in that the
physical/biological/chemical mechanisms that cause lung cancer have not been convincingly related to
mesothelioma mechanisms.
Section 7
1) The Aeolus model that lumps amosite and crocidolite into one amphibole category must be re-
evaluated. The fiber size distributions of the two amphiboles are extremely disparate, as seen in the
tables and figures of Appendix B. For example, crocidolite has a lower PCME, -0.007, versus -0.3 for
amosite. Chrysotile and the minor amphiboles are midway at -0.04-0.12.
2) Equations 7-1 and 7-4 use the percentage of PCME fibers as denominators of lung cancer potency.
This reduces the significance of fibers shorter than 5 jim and/or narrower than 0.25 |im. A review of
the raw data in Appendix B reveals that using PCME as a denominator disregards the role of more
than 90% of the airborne asbestos fibers. This extreme bias drives the modeled potency of these
numerous small fibers to be self-fulfillingly minor. This also creates positive bias for amosite potency
when one considers that the PCME for amosite is much greater than for other asbestos types.
Apparently a similar approach was used for mesothelioma, where downplaying these small fibers may
be an even more egregious oversight. Evidence continues to build that smaller fibers are more likely to
reach extrapulmonary sites and be associated with mesothelioma (Dodson et al., 1990; Dodson, 2006;
Suzuki & Yuen, 2001).
3) Yes, I agree with EPA 2003 decision, as outlined in Section 7.8, to not pursue the Aeolus method.
References
3-83
-------
Dodson, R.F. etal. Asbestos content of lung tissue, lymph nodes and pleural plaques from former
shipyard workers, Am. Rev. Respir. Dis. 142, 843-847, 1990.
Dodson, R.F. Analysis and relevance of asbestos burden in tissue. In Asbestos: Risk
Assessment, Epidemiology, and Health Effects, Dodson, R.F and Hammar, S.P. ed., CRC Press, Boca
Raton, FL, 2006.
Suzuki, Y., and Yuen, S.R. Asbestos tissue burden study on human malignant mesothelioma, Ind.
Health,39, 150-160,2001.
Charge Question 3c: For lung cancer, the current risk model is multiplicative with the risk from
smoking and other causes of lung cancer. Should the nature of the interaction between asbestos and
smoking be investigated further? If so, how should this be done? Do you think the model would be
sensitive to additional quantification of the interaction between smoking and asbestos?
One possible investigation would be the potential interaction of fiber-width exposure and smoking.
Fiber width is the primary determinant of aerodynamic diameter, which causes thicker fibers to be
intercepted in the upper portions of the bronchial tree. The compromised mucociliary escalator in
smokers probably enhances residence time of these thicker fibers in this upper region. Hence an
epidemiologic study using fiber-width bins for asbestos-worker exposure might reveal different lung
cancer potencies for specific fiber widths for smokers versus non-smokers.
Charge Question 8a. Do you agree that multiple binning strategies should be evaluated, or do you
believe that a physiological basis exists that can be used to identify a particular set of length and width
cutoffs that should be assessed? If so, what would those length and width cutoffs be, andean these bins
be implemented considering the limitations in the available TEMparticles size data sets? (see Section
10)
Multiple binning should be evaluated, but only using TEM-analyzed environmental exposure data
that is directly associated with health outcomes. Studies continue to reveal the importance of fiber
width in potency. Fiber width is the most critical dimension in determining deposition site in the
respiratory system, plays a significant role in determining surface area exposed to tissue, and may be a
factor in mobilizing fibers from alveoli to pleural space. Future attempts to model fiber potency
should have at least two bins for width. One possible width division could be an aerodynamic diameter
of 2.5 jim, which is the cutpoint for EPA fine (-respirable) particles. This would be -0.5 jim width
for amphibole asbestos and -0.65 |im width for chrysotile.
Charge Question 11 a: Are the study-specific selection rules proposed above scientifically valid for the
intended uses? Should any additional selection rules be added?
The first criterion, "published in a refereed journal" is too restrictive given the paucity of studies that
have been suggested by OSWER for this undertaking. While publication in a peer-reviewed journal is a
good indicator of quality, one can't be certain that lack of such publication is an indicator of poor
quality. It would be appropriate to let a subset of the Science Advisory Board (SAB) act as reviewers of
unpublished candidate studies to determine if they meet certain quality criteria. Probably all members of
the SAB have served as reviewers for publications and would certainly wield a fine-toothed comb on
any studies that are proposed for inclusion. However, to retain the transparency of this effort, it would
3-84
-------
be appropriate to continue exclusion of those studies whose details cannot be disclosed to the public, as
suggested in Section 9.3.
Charge Question 12a: Are you aware of any studies that should be included in the model fitting effort
that are currently excluded or omitted? If so, what are these studies, and do they meet the requirements
for study inclusion?
Stayner et al.'s (2007) recent study, using TEM results from archived filters from a South Carolina
textile plant during the 1960s, provides the only set of data that can currently be used for reliable risk
assessment.
References
Leslie T Stayner, Eileen Kuempel, Steve Gilbert, Misty Hein and John Dement. An epidemiologic study
of the role of chrysotile asbestos fiber dimensions in determining respiratory disease risk in exposed
workers. Occup. Environ. Med. published online 20 Dec 2007;
Charge Question 13a: Is it scientifically justifiable to employ a default dust-to-PCM conversion factor
when there are no site-specific data available?
No. The OSWER statement in Section 10.2 "However, most values of CF are found to range between 1
and 10 PCM s/cc per mppcf (USEPA 1986)." is misleading when compared to the data presented in
Table C-l, where CF ranges from 0.1 to 21.9. Thus the potential for error in using any default value is
enormous when on-site impinger/PCM comparisons are not available. Impinger data cannot be reliably
used to derive PCM concentrations, let alone TEM fiber-size distribution.
Charge Question 14a: Are the point estimates and uncertainty distributions for the fraction amphibole
term proposed for each study scientifically valid?
No. The data from Addison and Davis for 81 samples of chrysotile were produced by XRD, which
cannot distinguish asbestiform tremolite from non-asbestiform tremolite. Furthermore, since XRD
measurements are made largely on the basis of mass, a single moderate tremolite fiber could weigh as
much as a ten, or even a thousand, chrysotile fibers. Given the unknown impact of any tremolite found
in these chrysotile samples, no amphibole fraction should be given to chrysotile that was not mixed with
amphibole asbestos on-site. But this skirts the issue that only TEM-analyzed environmental exposure
data that is directly correlated with health outcome should be used for risk assessment. TEM analysis
will unequivocally determine what fraction of airborne asbestos, if any, was amphibole.
Charge Question 14b: Is it scientifically valid to use surrogate TEM data to estimate bin-specific
concentrations and exposure values in studies where these data are not reported? If not, what
alternative approach could be followed, or what additional data would be helpful?
No. Comparisons of concurrently collected TEM and PCM concentrations have produced inconsistent
conversion factors. Hence, PCM data cannot be used to create TEM-equivalent fiber-size distributions.
3-85
-------
Charge Question 14d: Are the point estimates and uncertainty distributions for the fraction amphibole
term scientifically valid?
No. Summarizing my comments for 14.a: The tremolite contamination of chrysotile remians completely
indeterminate in a fiber-population sense. An amphibole fraction should not be applied to chrysotile.
Only TEM analysis of exposure environments can reliably determine the extent of exposure to
amphiboles.
Charge Question 14f: Would the model benefit by establishing a common lower cut-point in diameter to
normalize the lower detection limit across studies?
No. TEM analysts are capable of detecting even the thinnest asbestos fiber. Given the probable
importance of width in potency, raw diameter measurements from TEM should be utilized.
B. Need for Future Research
TEM Analsysis of Retrospective Environmental Exposure
The greatest need for the type of risk assessment outlined in the OSWER proposal is additional
environmental exposure data produced by TEM analysis. Only TEM can reliably reconstruct the
bivariate fiber-size distributions needed for modeling and unequivocally determine whether or not
amphibole asbestos was present along with chrysotile in the environments under study. Archived filters
representing occupational environments that produced lung cancer and mesothelioma should be sought
and analyzed by TEM, as done by Stayner et al. (2007). Even archived PCM slides from these
environments should be sought for TEM analysis. This would, of course, require development of a
methodology to transfer asbestos reliably from under the cover slip to a TEM grid.
Animal Studies
I leave it up to SAB panel members who have more expertise in this area to make recommendations.
However, I would strongly caution that non-inhalation studies be performed with fiber-size distributions
that are appropriate to the targeted cell types. Throwing big fibers onto pleural mesothelial cells does
not replicate what happens in people or animals, as evidenced by the works of Dodson (2006) and
Suzuki et al. (2001). Elutriation has proven to be an effective means of producing fibers of the desired
aerodynamic diameters (Webber et al. 2008).
References
Dodson, R.F. Analysis and relevance of asbestos burden in tissue. In Asbestos: Risk
Assessment, Epidemiology, and Health Effects., Dodson, R.F and Hammar, S.P. ed., CRC Press, Boca
Raton, FL, 2006.
Leslie T Stayner, Eileen Kuempel, Steve Gilbert, Misty Hein and John Dement. An epidemiologic study
of the role of chrysotile asbestos fiber dimensions in determining respiratory disease risk in exposed
workers. Occup. Environ. Med. published online 20 Dec 2007.
Suzuki, Y., and Yuen, S.R. Asbestos tissue burden study on human malignant mesothelioma, Ind.
Health, 39, 150-160,2001.
3-86
-------
Webber, J.S., DJ. Blake, TJ. Ward, and J.C. Pfau. 2008. Separation and Characterization of Respirable
Amphibole Fibers from Libby, Montana. Inhal. Toxicol. 20:8, 733 — 740.
3-87
-------
Additional Materials Supplied by Dr. Julian Peto
Are Attached
1985 Comments on the first draft of the 1986 EPA Report
1981 Report to the EPA on Asbestos in Schools
3-88
-------
*35~
Institute of Cancer Research: Royal Cancer Hospital
la anociatloo with
THE ROYAL MARS DEN HOSPITAL
DIVISION OF ETtDEMIOLOOY
CAHCQ Km«acM CUVAJOM Qua e* Emuaouer:
JULIAN PETO
BLOCK D
CLIFTON AVENUE,
SUTTON, SURREY
SM25PX
01 • 643 1901
Dr. Dennis J. Kotchmar,
United States Environmental Protection Agency,
Environmental Criteria & Assessment Office,
(MD-52),
Research Triangle Park,
North Carolina 27711, U.S.A.
Dear Dr.
13tb August.1985
chmar,
I enclose my comments on Dr. Nicholson's asbestos
report. As I have just written a similar report with Sir Richard
Doll in which we have discussed most of the contentious issues,
I have referred to our report (Doll and Peto, 1985) in several
places.
I have seen Sir Richard's comments, and agree with
all of them, but I have tried to avoid repeating them. I've
picked up a few spelling errors, but it might be worth getting
someone to proof-read the text again.
Yours sincerely,
Julian Peto
0005*4
-------
AIRBORNE ASBESTOS HEALTH ASSESSMENT UPDATE
EPA 600/8-84-003F
General comments
The report is well written, and provides an excellent
reference document. My main reservation, and one which is
inevitable in relation to any review on asbestos which attempts
to draw quantitative inferences on which policy decisions will
be based, is that none of the major questions can be answered
with much confidence.
Three important conclusions which I would not accept
as well established are:
(i) There is little evidence that crocidolite and amosite
are much more dangerous than chrysotile. There are no good
dose-response data for these amphiboles, but there is strong
*•
suggestive evidence that they may be very much more dangerous
than cbrysotile,.and this is now widely believed, particularly
"outside the U.S. Crocidolite may be particularly dangerous and
although little is used in the U.S. it would be irresponsible
to encourage its use elsewhere.
(ii) Asbestos causes a substantial risk of gastro-intestinal
and other non-respiratory cancers. This is not the main subject
of this report, but it has important implications in relation
to asbestos in water supplies. Several recent reports have
questioned the association (see for example, Doll and Peto, 1985),
and the issue remains unsolved.
0005*5
-------
(Hi) Exposure levels in asbestos-containing buildings, particularly
schools, are often high enough to cause a serious health hazard.
Measurements in Canada and Britain suggest very much lower levels
than have been reported in the U.S., and it is not clear whether
this reflects a real difference between the materials and the
way in which they were applied, or is due to differences in
methods of sampling and analysis. Asbestos removal is likely
to entail enormous costs and may even increase exposure, and
the reasons for the discrepancy between U.S. and other results
should be established before widespread removal is undertaken.
/
Chapter 3
(i) Selection of studies for dose-response estimation. The
exposure estimates for most studies are too unreliable to be
of much value. This is not important for the practical purpose
of estimating the dose-specific lung cancer risk for chrysotile,
V
as the geometric mean of the lung cancer risks presented as
the best overall estimate is very close to the estimates obtained
i
-in two chrysotile textile factories in which extensive measurements
•were made (Doll and Peto, 1985). For other exposures, however,
and particularly for crocidolite and amosite, this uncertainty
should be emphasised. Exposure estimates for studies used to
estimate the mesothelioma risk (Table 3-30, p.88) are all of
dubious reliability. There were no contemporary measurements
in three (Selikoff et al., Seidman et al. and Finkelstein), and
in the fourth (Peto) the mesothelioma risk was so high that we
subsequently concluded that a large proportion of the mesotheliomas
(we arbitrarily assumed 50%-, but this could be too low) were
0005«R
-------
due to the small amount of crocidolite used in the factory (Doll
and Peto, 1985).
(ii) Fibre type (see also General comments" above) pp. 103-113
The data in Table 3-35 show:
1) a consistently low ratio of pleural mesothelioma to
lung cancer for pure chrysotile
2) virtually .no peritoneal mesotbeliomas for pure chrysotile
3) a substantially higher ratio of pleural mesothelioma
to excess lung cancer, and a high peritoneal mesothelioma
risk,, in the "predominantly chrysotile11, "predominantly
crocidolite" and "mixed exposure" studies.
One obvious explanation of these data is that even quite
small amounts of crocidolite or amosite can cause a substantial
mesotbelioma risk, and that chrysotile alone almost never causes
peritoneal mesothelioma. Moreover, brief intense exposure to
chrysotile has never been shown to cause either lung cancer or
mesothelioma, in contrast to both amosite and crocidolite. These
' observations are dismissed as being of less importance than the
large differences in dose-specific risk between different industries
using the same fibre type, but this may be a very dangerous
assumption. For example, if the substantially higher mesotbelioma
risk in the "predominantly chrysotile" studies is due to amphiboles,
which constituted only a few per cent of the total asbestos
exposure in these studies, the dose-specific risk (at least for
mesothelioma) might, well be ten times higher for amphiboles than
for chrysotile. There are no good exposure data on which to
000587
-------
4.
base dose-specific risk estimates for either amosite or crocidolite,
but both seem to be more dangerous than chrysotile when used
in the same way. • .
Table
One aspect of/3-35 that deserves special comment and
explanation is the adjustment of the lung cancer excess among
Canadian chrysotile miners, from 46.0 to 126.2. This is the .
largest of the pure chrysotile studies and is often cited as showing
that the ratio of mesothelioma risk to excess lung cancer is
quite high for pure chrysotile. This adjustment greatly reduces
this ratio, and hence adds weight to the argument that chrysotile
is very different from the amphiboles.
(iii) Gastro-intestinal and other cancers pp. 94-97. See "General
comments" above. In our report (Doll and Peto, 1985) we commented
that the excess SMR for other cancers followed much the same
pattern as for G.I. cancers, being highest in cohorts that suffered
a large excess of lung cancer and mesothelioma. We inferred
(a) that there is no reason to single out G.I. cancers for special
"mention, and (b) that a general increase in all cancers seems
biologically implausible, and these increases may therefore all
• *
be due largely or even entirely to misdiagnosed lung cancers
and mesotbeliomas. The discussion of these issues on pp. 94-97
reaches a different conclusion, but it is well documented, and
my only specific comments are (a) our alternative' interpretation
might be mentioned and criticised, and (b) the statement on p. 97
that "the excess at other sites is ....... generally less than
G.I. cancer" does not seem to be supported by Table 3-33.
-------
5.
Chapter 5 t
•
The data on ambient asbestos exposure given in Chapter 5
stand in marked contrast to British data collected by the Department
of the Environment and cited in Doll and Peto (1985). The
British data suggest that ambient levels in contaminated buildings
are rarely much higher than 0.001 fibre/ml, while the U.S. data
include many very much higher values assuming a conversion factor
of 30 ug/m3/f/ml (p.136). This raises two important questions,
both of -which should be discussed.
1). The method of counting respirable asbestos fibres
.•
by electron microscopy used in the British study is
claimed to provide counts that are directly comparable
with optical microscope measurements. The major
difference between this method and that used in the
U.S. may be the dispersion of fibre clumps by ultra-
sonification (p. 138). Results in nanograms cannot
\
be converted to counts . of respirable fibres of a
specified size range reliably (pp. 154-156), and the
i .
report should either recommend the British method or
discuss its deficiencies.
2). Comparative studies of British and American measurements
are now being conducted by the British Department of
the Environment and the E.P.A., in an attempt to
resolve this issue. Preliminary results of this work
should show whether there is a real difference between
British and U.S. schools, or whether the measurements
taken in Britain or the U.S. (or both) were defective.
Can this information be obtained?
-------
Optical fibre counts are extremely unreliable at low
concentrations, as most of the fibres counted are not asbestos.
The results shown in Table 5-8 may therefore be very much too
high, particularly the first four, which are 100 times higher
than typical ambient measurements by electron microscopy.
Chapter 6
There is a striking contrast between the ratios of
mesothelioma to excess lung cancer shown for pure chrysotile
in Table 3-35 and the predicted ratios shown in Table 6-3. For
exposures of up to 20 years duration the male ratio in Table
6-3 is about 1:1 for exposure beginning at age 20 and about 1:2
for exposures beginning at age 30, while the overall ratio for
pure chrysotile in Table 3-35 is only about 1:10. This is because
the four cohorts used to estimate the mesothelioma risk all had
quite'.high mesothelioma rates (Table 3-30, p.88), and all were
exposed to some amphibole. It should perhaps be mentioned that
the mesothelioma*predictions are likely to be substantially too
"high for chrysotile. Conversely, however, it might also be
mentioned that there are no good dose-response data for amosite
or crocidolite, and the estimates shown in Tables 6-1 to 6-3
could be substantially too low for both lung cancer and mesothelioma
for arophiboles (see comments on Chapter 3).
»•
The data on household contacts (pp. 162-165) are important,
but it seems likely that they resulted from quite high concentrations
of respirable fibres, and in the absence of fibre counts in these
homes the inference that low-dose effects may be grossly underestimate
-------
by extrapolation (last sentence, p.164) should perhaps be deleted.
It seems extraordinary that the prevalence of abnormalities in
household contacts (35%) was almost as high as in asbestos
workers (45%; Table 6-5).
Minor points
p xii, 1. 9-10 : It might be better to say: "The risk of mesothelioma
is approximately proportional to cumulative exposure, and
also increases sharply with increasing time since first
exposure."
p xii, last 2 1 : "Uncertainties in conversion between optical
fiber counts and electron microscope fiber counts" might
be preferable. (See comments above on Chapter 5. If these
are accepted, many references to mass measurements will
have to be altered or qualified.)
p xiii. My address is now: Section of Epidemiology, Institute
of Cancer Research, Sutton, Surrey SM2 5PX, England.
p 2, 1. 1. Should fibre differences be dismissed? See comments
above.
p 2. 1.26. "Document"misspelled.
p 2. last 1. Replace "mass determinations "by" fiber counts"
p 10 British standard is now 0.5 fibers per ml.
p 46 last line. Replace 'are1 by 'may be1. In many studies,
errors in dose estimation may well exceed statistical
errors in response.
p 47 Equation 3-3d. It might be preferable to replace IL/IE
by relative risk, and replace f.d by the average cumulative
dose of the cohort.
-------
8.
p 48 1. 24. Insert "per" before f-y/ml. The units of k, are
(f-y/ml)"1.
Table 3-10. Peto 1980 could be replaced by the estimate
cited by Doll and Peto (1985) of kL x 100 - 0.54.
(Entire post-1932 cohort. The separate estimate for post-1950
•workers was 1.5.) Alternatively, as the paper from which
this is taken is still in press, a footnote could be added
to the table to avoid having to alter the text and fig.
3.7.
pp 54-57. Doll and Peto (1985) could be mentioned in a footnote,
'as it discusses most of these points in detail, and presents
updated mortality results for the Rochdale factory in a
larger cohort which includes short-service workers.
pp 63-64. I believe that the average duration of mining exposure
of Asbestos and Thetford residents is higher than that of
McDonald's cohort, which included many short-service workers
• who left the area.
»
pp 81, last sentence. Another possible explanation for the low
mesothelipma risk beyond 50 years in U.S. insulators is
that early recruits suffered less exposure, perhaps because
of different materials or work practices in the early 1920 's.
p 98 last para. 1. 4 "Eradication" misspelled.
p 105. First sentence. Difference in rel. risk (pre v post 1950
at Rochdale) no longer significant (0.05 < p < 0.10).
An important point : heavy brief chrysotile exposure has
never been shown to cause increased risk, unlike amosite
(Seidman et al., 1979) or crocidolite (Jones et al.)
p 111. para. 2 1. 8 (twice) and 1.10 "Peritoneal" misspelled.
-------
p 114. 1. 14. Peritoneal, not parietal.
p 163. Chi-squares in Table 6-5. The first (7.1) has been
calculated with no continuity correction, and the second
(114) seems slightly too high.
p 164. last para. " and adjusted to a continuous rather than
day-time exposure". This seems to imply that Tables 6:1,
6.2 and 6.3 were calculated for daytime exposures, but the
rates in these tables are stated on p. 157 to be for
continuous exposure, which is presumably 24 hours/day,
7 days/week. This should be clarified both on p. 157 and
on p.164. In particular, assuming continuous exposure at
.•
0.01 f/ml must exaggerate the likely risk, as few, if any,
individuals are likely to spend their entire lives in such
conditions.
p 168 para 2. 1. 13. "Tract" misspelled.
00059*
-------
AN ALTERNATIVE APPROACH FOR THE RISK ASSESSMENT
OF ASBESTOS IN SCHOOLS
Report to the U.S. EPA
Julian Peto
April 6, 1981
-------
TABLE OF CONTENTS
PAGE.
1.0 Description of Risk Assessment Methodology 1
1.1 Mesothelioma. . ~. -. j
1.2 The Effect of Exposure Level and Duration of-
Exposure 1
1.3 Lung Cancer 5
1.4 Exposure Measurements and Dose Relationships -5
1.5 Final Dose-response Formulae. . . _ ......... 6
1.6 Definition and Calculation of Lrfelong Risk 7
1.7 Choice of Death Ra'tes for Lung Cancer and for the
Calculation of s(a) . . . 7 ............ 7
2.0' Example Calculations
2.1 Calculation of Mesothelioma Risk
2.2 Calculation of Lung Cancer Excess Risk
3.0 References
-------
LIST OF FIGURES
FIGURE PAGE
1 Data from Selikoff and Peto, 1981; unpublished
and Henderson and Peto, 1981; unpublished 2
2 Data from Doll and Peto (1981) ...... 8.
3 Data_frora Doll and Peto (1981) 10
.•-..:..:-..-- : -- - 11ST OF TASKS
TABLE PAGE
» .. . -
1 Incidence,rates calculated from the formula
T -(T-T ) --.-.. A
o _ - ...
2 Survival rates for males and females, all races
(U.S. 1972) 11
3 Lung cancer death rates per 10 per annum 12
4 Calculation of excess mesothelioma risks . 13
5 Calculation of excess lung cancer risks 14
ii
-------
REVIEW OF ASBESTOS IN SCHOOLS EXTENSION
1.0 Description of Risk Assessment Methodology
The only dose/time relationship for a human cancer that. has been
examined in any detail is that of lung cancer in continuing cigarette
smokers and lifelong non-smokers (Doll, 1978). Lung cancer incxdence
!„ in non-smokers satisfies the relationship:
N
4 5
(i) I
while the excess I£ in contianiag saoteis {l3iDse ^» have smoked at
a constant rate continuously) as approximately*
4.5
(ii) I- = kj-. (duration of smoking) ' .
.The constants Ic, (smoking effect) and k^ (background independent of
smoking) determine the absolute risk. Kj. depends on the amount
smoked, inhalation and tar level; it is approximately 100 x k^ in
smokers of about a pack per day.
Ill Mesothelioma
Mesothelioma rates behave in a similar way. The incidence I£
following asbestos exposure JSelikoff and Peto, 1981; unpublished)
is:
, 3.5
(iii) Ip = kp. (time since first asbestos exposure;
while !„, the incidence in the unexposed population (Henderson and
Peto, 1981; unpublished) is: " ~~
(iv) I = k '
For" mesothelioma among U.S. insulation workers, the ratio k^fcjj is
approximately 10,000. (See fig. 1) The only marked difference
between -the models for asbestos^induced mesothelioma and smoking-
induced lung cancer relates to the effects of stopping exposure.
Lung cancer incidence remains approximately constant when smoking
stops (Doll, 1978) and its age-distribution is very different for
continuing smokers and ex-smokers. This appears not to be true of
mesothelioma, however, and equation (iii) fits the incidence pattern
in various studies irrespective of fibre type or duration of exposure
(Selikoff and Peto, 1981; unpublished) .
1.2 The Effect of Exposure Level and Duration of Exposure
A possible explanation of the preceding observations is that asbestos
initiates the process of mesothelial carcinogenesis, and that the
probability that one day's exposure will give rise to a mesothelioma
T years later is roughly proportional to T . Adding up the separate
effects of each day of exposure then leads to the prediction that
-------
ID'1
»»•***»•*
-*. -v i /'. 4_» . .
I LOG. SCALE)
10 ~ -
JO
KGRT ALIT Y AMDN~G" I N S ULAT I ON
WORKERS, PLOTTED AGAINST
TIKE -S^NCE- F.IRST-.AS3ES7DS ,
'EXPOSURE * :; '"''
y
\s
MORTALITY IN THE UNEXPOSED
GENERAL POPULATION X 1000,
PLOTTED AGAINST AGE*
2-t-
50
I
60
• •
70 60
50
YEARS (LOG SCALE)
AGE (LOS ANGELES, WITH NO ASBESTOS EXPOSURE)
TJHE.S:J,:CE FIRST.ASHESTOS;"EX?OSURE (INSULATORS)
. ;"- *TK:; SLOPE- OF -BOTH- Lr.;zs i? -APPROXZHATSLT'--^.^" "
FIGURE 1. Data from Selikoff and Peto, 1981; unpublished
and Henderson and Peto, 1981; unpublished
-------
continuous exposure will produce an incidence pattern that rises as
(time since first exposure) , brief exposure will produce an incidence
proportional to (time since first exposure) , and intermediate
duration produces an intermediate effect. More formally, this model
predicts that an exposure of duration T years will produce an
incidence proportional to T -(T-T ) at°time T years after first
exposure. The corresponding incidence patterns are shown in Table
1. These rather specific predictions cannot be tested directly.
The incidence patterns shown.in/Table 1 increase as (Time) following
1 years' exposure and as (Time) "for continuous exposure, but even
these extremes are both quite adequately described by.the approxima-
tion of equation (iii) that incidence rises as (Time) " irrespective
of duration of exposure. - : ^-
For the purposes of the present calculation, the predictions in
Table 1 carry two other important implications. (1) The predicted
risk is approximately proportional to duration of exposure for
durations of between 1 and 6 years. Movement of students between
schools can therefore be ignored, as the number of mesotheliomas in
2,000,000 people each exposed for 3 years will be roughly the same
as that among 1,000,000 exposed fojc 6 years. (2) The incidence
caused by 6 years' exposure 30-40 years after first exposure is
roughly 0.5 times that caused" t>y continuous exposure, and the corres
ponding factor for 10 years' exposure is about 0.7. Observed incidence
rates in cohorts of industrial workers who have suffered prolonged
exposure and whose exposure levels have been estimated can therefore
be used to develop a dose-response relationship which can be adjusted
for the effects of shorter exposures at measured levels.
This model, under which each increment of exposure produces an
additional independent increment in subsequent cancer risk, predicts
that dose-response is linear; in other words, the incidence of
raesothelioma, and hence the lifelong risk, will be proportional to
the fibre level following a given duration of exposure.
These relationships are summarized by the approximate formula for
excess mesothelioma incidence:
~ 35 ~ -
(iv) I_ = k-(time since first exposure) * -(fibre/ml),
the constant k being proportional to the product of average hours of
exposure per day and the duration adjustment factor discussed above
(0.5 for 6 years' duration, 0.7 for 10 years, 1.0 for continuous
exposure).
It would of course be possible to base predictions on the specific
model tabulated in Table 1, which would give much the same results
as. equation (iv)-,:but I prefer to use .it-only to estimate the dura-
tion adjustment factors. The effect of different durations of
exposure cannot be estimated accurately from any existing study, and
some speculative assumption has to be made for this purpose. The
advantages—of -using equation- ^iy) are -that if alternative factors
are suggested they can be substituted directly; and that it is
scientifically less, tendentious.. . - . . . - : _.-.--
. 3
-------
4 4
TABLE 1. Incidence rates calculated from the formula T -(T-TQ) , where T is
years since first exposure to asbestos, and T is duration of exposure.
Incidence rates among U.S. insulation workers first exposed 1922-1946
(per 10 per annum) are also shown.
Duration of asbestos exposure (T ) Incidence x 10
1 6 10 20 Continuous per annum (13.5. Insulators)
Years
since
first
*
expo-
sure
(T)
12
22
32
42
52
.5
.5
.5
.5
.5
2
11
35
80
151
6
49
167
400
786
7
62
231
578
1166
7
69
294
809
1743
•- 7
69
300
878
2044
61
277
.
647
. 1156
r
-------
In an earlier analysis (Peto, 1979) a quadratic raiaci. ,..
dence time model was analysed, and overall incidence was approxi-
mated by (time) rather than (time) * , but the estimates presented
here provide a better fit to the data now available. The exponent
of time cannot be estimated precisely, however. Even the estimate
based on 138 mesotheliomas (Selikoff and Peto, 1981; unpublished)
has a standard error of ±0.4. • -,
.3 Lung Cancer
The relative risk for lung cancer caused by prolonged exposure to
--asbestos appears to reach a maximum about 25_to 35 years after Jirst^
exposure. -In-one- study-we have conducted~lh"Englanrd7""and" in North
American insulators, (Sellkoff -et al., 39793-. ^e relative risk
subsequently fell, birt this decline wary be an-artifact. In both
these studies the relative risk reached a very high level (about 6
in American insulators, (Selikoff et al., 1979), and more than 10 in
English asbestos textile workers heavily exposed before 1933). This
-high mortality, together with deaths due to asbestosis and mesothelioma
may have selectively removed the most heavily exposed men, particular-
- ly among the heavier cigarette smokers. It is therefore reasonable
to suppose that the relative risk would reach a maximum about 20
years after first exposure and subsequently remain at about this
level following less prolonged exposure at lower dust levels.
If asbestos is eventually inactivated or eliminated, the effect of
brief exposure in childhood could be negligible. On the other hand,
there is some evidence that the eventual relative risk following
both brief and prolonged exposure increases slightly as age at first
exposure falls (Selikoff et al., 1973), and it could be argued that
the effect of childhood exposure might be greater than among those
first exposed as adults. The assumption that the relative risk will
•rise for about 20 years and then remain constant following exposures
of up to 10 years duration, and that the eventual proportional
excess (i.e. relative risk minus 1) will be proportional to both
dust level and duration of exposure but independent of age, seems a
reasonable compromise between these extremes. Thus for lung cancer
the calculation will be based on the formula for excess incidence
I.£.GO__at _age(aj_: • '
i
(v)-- IE(a) = c- (fibre/ml) • (duration of exposure)- I|j(a)»
where I.-(a) denotes the lung cancer incidence in an individual of '
age(a) who has not suffered asbestos exposure. As the effects of i
smoking and asbestos are approximately multiplicative for lung
cancer the risk calculation must be carried out separately for
smokers and non-smokers, using appropriate- age, sex and smoking- '
specific lung cancer rates I,,(a). i
1.4 Exposure Measurements and Dose Relationships
i
Methods for measuring fibre counts have altered in recent years, the,
most significant change being the use of a microscope eyepiece
-------
graticule in fibre counting (Steel, 1979). This procedure increases
the fibre count by a factor of 2 or more, and throughout
the following discussion fibre counts are assumed to have been
measured in this way. Thus, for example, the old estimate of tne-
average exposure of North American insulation workers exposures, 10
to 15 fibre/ml (Nicholson, 1976), becomes about 30 fibre/ml.
uttH fibre/ml) -.(duration of exposure) in equation (v) is
simply cumulative dose, and the constant c can be estimated from
data on industrial cohorts such as~ the North American insulators.
In view of the delay of about 2$ years before the maximum relative
risk for lung cancer is reached, however, recent exposure will not
have had its full effect. - *t Is -difficult :*o justify, any very pre-
cise adjustment to 'allow for this -"wasted exposure". A reasonable
approximation migh be to "assume that the relative risk of about 6 (a
proportional excess risk of about 5) among North American .insulators
30-35 vears after first exposure reflects the effects of 20 years -
exposure at 30 fibre/ml, or 600 fibre/ml years. The; constant c can
then be calculated from equation (v) : - - - --- . ....
(vi) c = (relative risk minus I)/ (cumulative dose)
= 0.0083,- i_: ---: -r: •.. .----- .,
since I (a)/Ir(a) is by definition relative risk minus 1.
Finally, some adjustment must be made for the shorter working week
and longer holidays of school pupils and staff compared with in-
dustrial workers. A factor of 0.5 will be assumed for both, giving
a final estiaate for. schools for the constant c of 0.0042.
For mesothelioma, the constant k in equation (iv) ' can also be esti-
mated from the experience of North American insulators. Their
incidence of mesothelioroa 30-35 years after the start of prolonged
exposure at about 30 fibre/ml was 3 x 10 per annum (Selikoff and
Peto, 1981; unpublished), so
(vii); k = 3 x 10"3/{32.53'5.30)
..... ____ _________ _____ . ____.e .1 -, , "i A .._. __ ._ _________________ ..... ____________
—3.1 X 1U .
The constant must be adjusted For the shorter working week in schools
(a factor of about 0.5, as above), and for duration of exposure (a
further factor of 0.5 for 6 years' exposure or 0.7 for 10 ye|£S
exposure; see above). Thus k =5.1 x 10 x 0.5 = 1.3 x 10 (6
years' exposure), or 1.8 x 10 - (- 10 years' exposure.
1.5 Final Dose-response Formulae
Lung cancer 20 or more years after first exposure:
(viii) Excess incidence = normal (age, sex and smoking-
_._ .specific) incidence x fibre/ml in school x duration
........ "ri:x 0.0042. -: --: : -:::-::: .-.-.-..-.
-------
The excess during the first 20 years will be so much lower than
later, due to the lower initial relative risk, and to the steep
increase with "age of lung cancer incidence in both smokers and
non-smokers, that it can be ignored.
Mesothelioma:
3.5
(ix) Excess annual incidence .= k-(years since first exp.)
x (fibre/ml in school),
where k = 1.3 x 10*^ (6 years' exposure), or
k = 1.8 x 10~ (10 years* exposure).
1.6 Definition and Calculation of Lifelong 3tlsk
Lifelong risk is defined as" the probability'that an individual
exposed to asbestos will die as a result of his exposure. The
calculation of lifelong risk depends only on the age-specific survival
rate's(a) of the population at each age(a), and the age-specific
excess incidence I(a) caused by asbestos for each disease. (Incidence
and mortality are similar for both lung cancer and mesothelioma.)
The lifelong" risk is then simply the product of s(a)'I(a) over all
ages greater that a , divided by s(a ), where aQ is age at first
exposure. Deaths occurring after age 80 are ignored in the following
calculations. They can of course be added if required, but this
would entail extrapolation well beyond any existing observations,
particularly for mesothelioma, and would not greatly affect the
results. It is assumed throughout that individual risks due to
exposures of the sort encountered in schools are too low to appre-
ciably affect life expectancy, even if•the number of resulting
deaths is substantial.
1.7 Choice of Death Rates for lung Cancer and for the Calculation of s(a)
It is difficult to predict overall mortality in the future. For the
purpose of predicting mesothelioroa mortality the life-table based on
1972 U.S. national mortality, combining all races and both sexes,
has been used (Table 2). Death rates have already fallen substantially^
since 1972, however (Fig"."2)",~ana are' liKeiy to fall further.- Tirts
will increase the lifelong risk due to mesothelioma, perhaps substan-
tially, as the probability of surviving to old age, when the meso-
thelioma risk is highest, will increase. These calculations should
be repeated using projections of future national death rates, if any
are available.
For the lung cancer projections for smokers and non-smokers the
position is still more difficult. Female lung cancer rates in old
age are very much lower than male rates but if current smoking
trends continue the sexes may well suffer similar rates 50 years
hence at all ages. The effects of future changes in tar level and
consumption may also be considerable in both sexes. U.S. male lung
cancer rates are still rising at all ages above 50 due to the "smoking
cohort" effect, whereas in Britain this has now ceased, and British
-------
FIGURE 2 Data from Doll-and Peto (1981)
DECREASING
AOULT AND
ESPECIALLY.
CHILDHOOD
MORTALITY"
DECREASING NON-RESPIRATORY CANCER, f EI4ALES
; * —- .
r NON-RESPIRATORY CANCER. MALES
=*=
MHQ IMS IKO 1955 I960
CENTRAL YEAR
DECREASING
VASCULAR DISEASE
SrNCE 1B7O
t»65 1870 IB7&
fcrnu.l r«r 0T 5-y««r p*iiod nud-»iJ. 1833-37 u> 1B73-77)
-Annual agr-standardized death rates, 1933-77, among Americans under 65 yean o( age.
-------
rates are falling slightly between ages 50 and 65, and markedly
below age 50 (Fig. 3), probably due largely to the switch to filter
cigarettes in about 1960 in Britain. It is clear from Fig. 3 that
U.S. male rates are likely to exceed British rates at all ages
within a decade or so. They are already equal or higher below age
55, and at older ages the curves are converging rapidly.
The lung cancer projections for non-smokers given below are based on
the U.S. male death rates for lung cancer and all causes from Hammond
(1966) (Tables 2 and 3). For smokers, rates for all causes in male
current smokers have also been taken from Hammond (1966) (Table 2)
but current British male lung cancer rates, inflated by a factor of
50%.at each age to-allow J:ojr ±he .exclusion of non-smokers and ex-smokers,
have_ been used as a basis for projection of future .TJ-S. lung cancer
ra'tes among male current smokers ITV the T3.5. (Table 3). For the
reasons outlined in the preceding paragraph these would provide a"
better basis_for prediction than Hammond's 1966 rates for current
smokers if current smoking habits persist. This constitutes a
necessarily arbitrary guess, however. If either tar levels or
cigarette consumption fall substantiaTly~iETtb.e~ future the resulting
predictions will be too high, and the calculation and corresponding
predictions must be regarded as illustrative rather than definitive.
More recent American Cancer Society data should be used if available.
No predictions of female lung cancer excesses are presented. For
the reasons outlined above, it is possible that the risk to women
will be similar to that among men, but no firmer statement can be
defended.
Apart from the difficulties outlined above and the fact that the
calculations for mesothelioma are not based on survival rates of the
same population as those used for lung cancer, survival rates for
male smokers and non-smokers (column's (B) and (C), Table 2) vere not
available below age 35. (See notes to Table-2). These irritating
details are mentioned to emphasize again-that the following calcula-
tiojtis should be regarded as illustrative of the method, and do not
provide results which can be quoted out of context.
2-0—£xampl€-Galculations-
2.1 Calculation of Mesothelioma Risk
Using column (A) in Table 2 as the survival rate s(a) at age(a) and
computing annual excess mesothelioma incidence I(a) from equation
(ix), the calculation of risk following exposure at 1 fibre/ml for 6
years starting at age_a = 12 (k = 1.3 x 10* ) or 10 years from age
a = 30 (k = 1.8 x 10 ) is shown in Table A. The number of deaths
per 10 individuals exposed from age a in each 5-year age-range = 5
x 10 x I(a) x s(a)/s (a ). The risk caused by any other concentra-
tion is calculated by simple proportion. Thus, for example, at
0.002 fibre/ml the lifelong risk would be 0.002 x 329.1 =.0.66
deaths per 10 exposed children. Note that most deaths (71% in
children, and 85% in adults) would occur after age 60.
-------
FIGURE 3 Data_frora Doll and Peto (1981)
-Comparison oE lung cancer trends in the U.S.
(USA, round symbols) in selected age groups with corresponding
trends in England and Wales (EirW, square symbols).
Data from tables E/-EY; points for 1940 to 1950 arc estimates
corrected for under-ccrtification (open lymbols); points for subse-
quent years are observed rates (solid symbols).
10
-------
TABLE 2. Survival rates (A) for males and females, all races (U.S. 1972);
(B) for male non-smokers; and (C) for male current smokers of 10-19 cigarettes
per day. (B) and (C) are calculated from Hammond (1966) Appendix Tables 2a
and 3a. Hammond (1966) did not present death-rates for lower ages, and the
survival rates at lower ages have been arbitarily assumed to be the same
for (B) and (C) as for (A).
Age
(a)
0
12
22.5
27.5
, 30
32.5
37.5
42.5
47.5
52.5
57.5
62.5
67.5
72.5
77.5
Survival s(a)
(A)
Total U.S.
1 . 000
.976
.966
.959
.955
.951
.942
.928
.907
.876
.831
.760
.679
.569
.432
Survival s(a)
(B)
Male NS
" 1.000
.976
.966
.959
.955
.951
.942
.934
.923
.911
.886
.849
.783
.690
.560
Survival s(a)
(C)
Male S
1.000
.976
.966
.959
.955
.951
.942
.932
.917
.885
.836
.765
.660
.523
.387
11
f
-------
TABLE 3. Lung cancer death rates per 10 per annum (A) for non-smokers
(Hammond, 1966) and (B) for male smokers. The smokers' rates (B) were
estimated by.multiplying national British rates for 1977 by a factor of
1.5.
Age
Ju(a)
. . . Non-Smokers-- .
(A)
, Va)
Smokers - - ~
(B)
" • - : . . : :
32.5
37.5
42.5
' .. 47.5
52.5
57.5
62.5
67.5
72.5
77.5
X"
-
2.3
5.0.
4.9
10.5
13.9
14.7
16.1
35.8
2.*
8.1
20.1
63.8
159.9
273.3
500.1
763.7
1010.4
1197.5 "~
12
-------
TABLE 4. Calculation of excess mesothelioma risks in successive 5-year age
intervals up to age 80 due to exposure at 1 fibre/ml for 6 years from age
12, or 10 years from age 30.
Age
(a)
0
ao=12
20-24
25-29
a = 30
0
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
75-79
Survival at
mid-point
of interval
s(a)
1.000 .
s(ao)=.976
.966
.959
s(aQ)=.955
.951
.942
.928
.907
.876
.831
.760 -
.679
.569
.432
Children
6 years from age 12
Incidence Deaths in
I(a)xlO"> 5-year -fn-
terval x 10
<0.1
0.2
0.5
1.1
2.0 .
3.5
5.5
8.3
11.9
16.6
22.4
29.6
0.2
0.9
2.5
5.3
9.7
16.1
24.7
35.2
'46.3
57.6
65.3
65.4
Staff
10 years from age
Incidence Deaths
I(a)xlO year in
xlO
-
<0.1
0.1
0.4
1.0
2.0
3.5
5.8
9.0
13.3
-
0.1
0.6
1.9
4.5
8.5
14.0
20.7
26.8
30.1
30
in 5-
terval
TOTAL DEATHS PER 10"
329.1
107.2
13
-------
2.2 Calculation of Lung Cancer Excess Risk
Using column (C) in Table 2 as the survival rate s(a) at age(a), and
column (B) in Table 3 as the lung cancer rate I«(a) in male smokers,
the corresponding risk calculations (see equation (viii)) are shown
in Table 5. The excess during the first 20 years after-first exposure
— is assumed to be negligible. The risk is proportional to duration
and to fibre level, but age at first exposure has virtually no
effect. A concentration of 0.002 fibre/ml for 6 years from age 12,
for example, would thus cause 0.002 x 301.6 = 0.60 deaths per 10
exposed children. As for mesothelioma, the majority of deaths occur
after age 60 (80% in children, and 84% in adults).
Repeating these calculations for non-smokers, using column (B) in
Table 2 and column fA) in Table 3 gives predicted total excess lung
cancer risks per 10 at 1 fibre/ml of 9.6 (compared with 301.6 in
smokers) for children and 15.0 (compared with 494.4 in smokers) for
adult staff. These are an order of magnitude lower than the predicted
mesothelioma risks, and for practical purposes can be ignored.
-------
TABLE 5. Calculation of excess lung cancer risks in successive 5-year age
intervals up to age 80 following exposure at 1 fibre/ml for 6 years from
age 12, or 10 years from age 30. (Male smokers of 10-19 cigarettes/day.)
Age Survival at
(a) mid-point
of interval
s(a)
0 1.000
a = 12 s(a )=.976
o o
'20-24 .966
25-29 .959
a = 30 s(a )=.955
•o o
30-34 .951
35-39 .942
40-44 .932
45-49
50-54
55-59
60-64
65-69
70-74
75-79
TOTAL DEATHS PER 105:
.917
.885
.836
.765
.660
.523
^387
Lung cancer Children Staff
death-rate 6 years from age 12 10 years from age 2
in unexposed -
smokers xlO Deaths per 10 in 5-year interval*
2.6
8.1
20.1
63.8
159.9
273.3
- 500.1
763.7
1010.4
1197.5
. 0.3
1.0
2.4
7.6
18^.3
29.5
49.4
65.1
68.2
59.8
301.6
-
-
31.1
50.2
84.1
110.8
116.2
101.9
494.4
30
LTTV
15
-------
3 v :'
. . 3.0 References ' - - - -
Doll R. 1978. An epidemiological perspective of the biology of cancer. Cancer
Res. 38:357.3^3583.
4
Doll R and Peto R, 1981. The causes of cancer: quantitative estimates of
risks.of cancer in America today. J. Natl. Cancer Inst. In press,
Hammond" EC. 1966.- -Smoking in-relation-to death ":rate&.-:"In:: - Na-tl.--Carkrer I-rist.
Monograph 19., -- ::_:•-:...-...
Henderson BE and Peto J. 1981. (unpublished)
Peto J. 1979. Dose-response relationships for asbestos-related disease:
implications for hygiene standards"?'Part II: Mortality. Ann. N.Y. Acad.
Scf. -330:195-203. - : -
i
Selikoff IJ, Hammond EC," and Seidman H.. 1979. Mortality experience of "
insulation workers in the U.S. and Canada, 1943-1976.' Ann. N.Y. Acad.Sci.
330:91-116.
Selikoff IJ, Hammond EC, and Seidman H. 1973. Cancer risk of insulation
workers in the U.S. In: " Biological effects of asbestos. I.A.R.C. Lyon. '
Selikoff IJ and Peto J. 1981. Mesothelioma incidence among asbestos workers:
implications for models of carcinogenesis .and risk assessment calculations
(unpublished) .
Steel J. 1979. Asbestos control limits. In: 'Asbestos. Final report of the
advisory committee. london: Health and Safety Executive, H.M.S.0., pp.85-87.
16
------- |