United States f—
Environmental Protection
Agency
Research and Development
APPENDIX TO
SELECTED APPROACHES TO
RISK ASSESSMENT FOR MULTIPLE
CHEMICAL EXPOSURES
-------
United States
Environmental Protection
Agency
Research and Development
APPENDIX TO
SELECTED APPROACHES TO
RISK ASSESSMENT FOR MULTIPLE
CHEMICAL EXPOSURES
-------
APPENDIX TO SELECTED APPROACHES
FOR MULTIPLE CHEMICAL E
Dr. Jerry F. Stara, Direct
Editor
Environmental Criteria and Asse
Cincinnati, Ohio 452
Dr. Linda S. Erdrelc
Technical Editor
Environmental Criteria and Asse
Cincinnati, Ohio 452
Contract No. 68-03-3111 by Dynan
Environmental Criteria and Assi
Office of Health and Environment
Office of Research and Dev
U.S. Environmental Protect'
Cincinnati, OH 452f
-------
NOTICE
This document has been reviewed 1n accordance with U.S. Environmental
Protection Agency policy and approved for publication. Mention of trade
names or commercial products does not constitute endorsement or recommenda-
tion for use.
A-li
-------
CONTENTS
Page
1. INTRODUCTION A-l
2. POSTMEETING MEMORANDA A-2
Andelman, Julian A-3
Bingham, Eula A-7
Calabrese, Edward A-12
Clarkson, Thomas A-l4
Cornish, Herbert A-l7
Crump, Kenny A-20
Durkln, Patrick A-23
Ensleln, Kurt A-26
Hartung, Rolf A-27
Hattls, Dale A-34
Legator, Marvin A-44
Mans on, Jeanne A-45
Mehlman, Myron A-47
Nicholson, William A-49
Nisbet, Ian A-52
0'Flaherty, Ellen A-56
Schneiderman, Marvin A-61
Silbergeld, Ellen A-74
Withey, James A-82
Wyzga, Ron A-88
3. REFERENCES CITED IN MEMORANDA A-92
4. BACKGROUND REFERENCES A-94
A-iii
-------
SECTION 1
INTRODUCTION
This Is the appendix to the report entitled "Selected Approaches to
Risk Assessment for Multiple Chemical Exposures." It contains the post-
meeting responses. In the form of memoranda, of workshop participants. The
participants were asked to comment on each consensus topic, pointing out
the extent of consensus and obstacles to agreement, and to elaborate on
topics of particular Interest that were presented or discussed 1n the
workshop sessions. Also Included 1n this appendix are a 11st of references
cited 1n the postmeetlng memoranda and copies of two background references
(Stara et al., 1981; Crump and Howe, 1983) that were distributed to the
participants prior to the meeting.
A-l
-------
SECTION 2
POSTMEETING MEMORANDA
A-2
-------
Julian B. Andelman
August 4, 1983
Postmeetinq Memorandum
Workshop of Risk Assessment for Multiple Chemical Exposure, EPA, Cincinnati,
Ohio, July 12-13, 1983
PART I. CONSENSUS TOPICS
A. Interspecies Conversion of Dose and Duration of Exposure
A considerable portion of the discussion was devoted to the question
of the power of the animal weight that should be used in comparing doses
between species (e.g., the one-third power) and similarly the power of
time (e.g., square root of time). Clarkson noted that the ratio tl/2/ml/3
seems to be constant among species for methyl mercury. There seemed
to be a consensus that such an approach was useful, but disagreement as
to what the exact mathematical relationship might be. There was no systematic
presentation of the options and sufficient experimental data to back them
up. It is likely that there are differences among chemicals and not a
constant relationship to use among all species. Thus, it may be important
to in fact develop more than one relationship. Therefore, I recommend
that an analysis of actual data and models be developed, and that, based
on it, a judgment made as to the possible range of relationships to be used.
It was not clear to me, based on the discussion, that an obvious single
relationship is applicable.
B. Health Risk Assessment for Less than Lifetime Exposure
Crump1s mathematical treatment is elegant and well-presented. One
question is whether it can be used in fact to distinguish between stage
of carcinogenesis and thereby be applied to assess whether an earlier
or later period of exposure was more important. Enterline noted that
his work on arsenic-induced lung cancer was not able to effectively
distinguish among stages by such a model. Another issue relative to the
Crump model is whether the concentration term (dose) can be effectively
dealt with when it is not a first-order variable (e.g., concentration to
A-3
-------
to the 1.2 power). This seems to be a not uncommon situation and should
be an essential part of the model. For example, in the EPA Water Quality
Criteria Document for Arsenic, the fit of the epidemiological data to
a model indicates such a non-linear dose factor.
C. API's Based on Quanta!, Continuous or Graded Data
The goal of this discussion should be to develop an approach which
would determine an ADI which would be protective of a defined percentage
of the population with a given probability or uncertainty. Thus, it
should not be to establish such a level protective at a 100 percent
certainty. Then, once such an ADI is determined, a safety factor (not
an uncertainty factor) can be incorporated. Presumably the uncertainty
factor will already have been incorporated into the ADI, initially.
On this basis, it should be decided as to which specific effects
are to be incorporated into the ADI. That is, should it include simply
esthetic effects, such as organoleptic ones or only frank disease
effects. The EPA Interim Drinking Water Regulation distinguish Primary
( health related) and Secondary (aesthetic and other) effects which
are regulated differently. The approach by Yves Alarie in his article
"Dose-Response Analysis in Animal Studies: Prediction of Human Responses"
should be considered. (Alarie, Env. Health Perspectives, 42, 9-13, 1981).
For sensory irritants he showed how the effects of respiratory rates in
animals could be used to predict varying intensities of responses in
humans.
D and E. Route-to-route Conversion and Mutiple Chemical Assessment
These topics are related and will be considered together. Durkin
presented the approach of summtng the exposure/ADI for each chemical
when a multiple chemical exposure is involved. As discussed, one
important factor is whether chemicals act in concert to cause the
same type of effect. That is, if they are independent in their actions
involving different target organs, such additivities of exposure/ADI
are irrelevant in terms of predicting when there may be a critical
level exceeded. A second factor to consider is whether the fact that
A-4
-------
the sum of such ratios happens to exceed unity should be a critical point
in deciding whether the population has been exposed to a critical quantity
of chemicals. This could be very misleading, again, if there is no
understanding of whether the effects are independent. It seems to me
that before such a "single number" approach is developed, some further
consideration should be given to it and, specifically, a presentation
developed which considers actual experimental data.
With respect to route-to-route conversion, we seem to be discussing
only broad principles here and have not yet considered a sufficient
number of examples to take us beyond the Stokinger-Woodward model.
It seems that there are two possible approaches to use in the absence
of detailed experimental data: first, to assume complete absorption
by both routes (oral and inhalation) and then simply predicting an
effect in one from the other. This would, however, not necessarily
be conservative, since there would be a chance, depending on the relative
amount of absorption by the two routes, for underestimating the dose
for the second route from the first. A second approach would be to
establish a range of relative exposures from an analysis of various
chamicals and their different routes, the example being the work of
Pozzani et al, in Amer. Industrial Hygiene Journal, 20:364-369 (1959).
PART II. NEW TOPICS
A. Structure Activity Relationships
Enslein discussed the use of structure activity, (S-A) relationships,
particularly to see if concordance could be shown between a qualitative
prediction of an effect and the S-A approach. The success for qualitative
predictability is impressive, but the real need here is for obtaining
quantitative relationships. Although the use of S-A was shown to
be very advanced in aquatic toxicity, it may be too early to use it for.
our purposes.
B. Reproductive Effects - No comments
C. Use and Biological Justification of Math.Models -
One important area seems to have to not been covered in this discussion,
A-5
-------
namely the need to deal with a population distribution in assessing
exposure and risks, as well as the distribution of chemicals in such a
population. Thus, even if there is a nominal constant concentration
in water or air, the behavior of the people will result in a distribution
of actual exposures which can be quite varied and should be factored
in to the analysis.
PART III. WORKGROUPS
A. 1 Sensitive subgroups
There is one issue not discussed which I would like to mention. There
may be sub-groups which are particularly sensitive to one agent as a result
of their exposure to another. For example, there is clear evidence that
incidence of non-melanoma skin cancer is related to exposure to UV-B sunlight.
There is also some evidence that arsenic exposure might promote this effect.
Such relationships could be considered in the framework of sensitive sub-
groups.
A.2 Ranking the Severity of Effects
It seems that a ranking scheme should only be developed and should
not be a "seat-of-the-pants" approach. That is, whether liver damage
is a more harmful effect than kidney damage should only be decided after
these principles are set forth.
B.1. Multiple Route Exposures
The question of multiple route exposure evaluation is more directly
considered under that of route-to-route extrapolation, at least in terms
of equivalency of effect. In terms of assessing actual exposures this is not
different than that of evaluating exposure via any given route. However,
I would like to draw attention to the growing evidence that off-gassing
of volatile chemicals from water used in the home can creat an air
exposure separately from that of the direct ingestion of the water.
B.2. Use of Exposure Data in Assessing Health Risk
As part ot the "writinq" team for that group, I have nothing further
to add.
A-6
-------
Postmeeting Memorandum of Eula Bingham
CONSENSUS TOPICS
I. Interspecies conversion of dose and duration of exposure
A. EPA approach
The approach by EPA in setting Water Quality Criteria has
3
been to use the factor mg/m in making interspecies conversions
of dose and the fraction of life span for the duration of
exposure (non-cancer effects)*
B. Consensus
2
It was the consensus that using the mg chemical/cm body
surface area per day was probably the best that could be done
and does offer advantages over mg chemical/kg body weight per
day. The only viable option currently for duration is using
fraction of life span.
C. Obstacles
The major obstacles are that none of these factors alone will
provide perfect equitoxic doses since the differences among
species cannot be accounted for by these simple terms, however
to be more precise requires more data on functional differences
not just more complicated mathematical modeling of an inadequate
data base.
II. Health risk assessment for less than lifetime exposures
A. EPA approach is to use actual studies for determining criteria
for aquatic organisms and the only human health calculation
involves dividing the subchronic exposure level by 10 to
estimate a corresponding chronic level, (reverse procedure
chronic —> subchronic has not been done)
a-7
-------
2
B. Consensus
It appeared to me that the group was in favor of using alj_
the data to calculate the best fitting NOAEL.
C. Obstacles
A major obstacle is the lack of data at various exposure
durations so it was considered the best recourse to use all
data in making any estimate.
III. Carcinogen
A. EPA approach
Currently the agency bases lifetime cancer risks from
durations less than a lifetime on the ratio of exposure total
dose to the total dose associated with a given cancer potency
and lifetime cancer level.
6. Consensus
The consensus was to use the multistage model (Crump and
Howe, 1983) described at the meeting by Dr. Crump.
C. Obstacles
Probably fewer than for most of the other consensus issues
but still data are missing to validate.
IV. ADI's based on quantal, continuous or graded data
A. EPA approach
The current approach by EPA is to use a safety factor
approach. The goal of this group was to select a method for
estimating ADI's using quantal or continuous data.
6. Consensus
It was not clear to me that there was a consensus.
A-8
-------
3
C. Obstacles and other
The obstacles are clearly the disadvantages raised by Dr.
Crump, mainly the nature of the toxic event should guide the
choice of the model. For effects other than genetic in origin
a linear dose response model may over estimate the true risk.
From a public health view this is not such a handicap but from
the view of establishing a cost-benefit analysis it may
compromise the agency's position. It is probably worthwhile
to use Crump's model to test a few chemicals (non-carcinogens)
about which we have data to determine the usefulness.
V. Route to route conversion
A. EPA approach
Currently it has been the practice of the Agency to use
inhalation data or TLV's to estimate oral route exposures using
the Stokinger-Woodward method.
B. Consensus
It seemed to me that generally the group agreed with the
proposition that Dr. Withey put forth that route to route
conversions be limited to inhalation and oral routes as well as
with the conditions he laid out as acceptable for extrapolation,
i.e. 1.) portal entry is not target organ, 2.) toxicant is not
totally inactivated prior to reaching target, and 3.) elimination
rates are sufficiently slow that blood levels do not fluctuate
greatly during dosing.
While I am in general agreement, it may be that Item 1 is always
necessarily required.
A-9
-------
4
VI. Multiple Chemical Assessments
A. EPA approach
It was not clear what the current approach is except that for
carcinogens the approach is to multiply or add up the risks
(with such small numbers there is essentially no difference).
B. Consensus
There was really only time to hear Dr. Durkin present the
issue and options.
C. Obstacles
Lack of time to discuss this issue was inhibiting. If I had
to select one option I would probably go with #2, but I am not
satisfied with many of the assumptions, e.g. two tumors in the
same organism are no worse than one tumor. The judgment runs
contrary to the way experimental data on the potency of carcinogens
has been judged by many investigators (Berenblum, Shubik et al.).
VII. Workshop - Considering high risk subgroups in health risk assessment
A. The question of how to account for sensitive subgroups in a
population is one that the group has discussed previously. During
this session this participant acted as the moderator. A series
of questions were posed to the group.
B. Consensus (Dr. Schneiderman took notes and will provide those.)
Generally the group agreed that it would be worthwhile to explore
the development of criteria defining specific high risk groups. It
seemed reasonable to come up both with lists of chemicals to which
there are hyper-sensitivities and to identify groups of people
who may be at special risk. A number of suggestions for how to
go about this were given (see individual reports).
A-10
-------
5
These include disease registries, estimates of populations
with various disease states, e.g. asmatics, heart disease,
renal failure, etc. Another suggestion was how much more
susceptible are children than adults to carcinogens using
animal and (human) data bases.
Biochemical, enzyme, and/or pharmacokinetic variations can
be used to reach certain conclusions regarding variability.
My impression was that there was consensus regarding the
usefulness and importance of bringing this information
together.
C. Obstacles
Major obstacles were finding the data in the literature or
within various agencies. It seems to me that one or two of the
participants can help with this problem.
VIII. Workshop on Ranking the Severity of Effects
A. Apparently the main issue in this workshop was whether or not
a "proposal" in the Federal Register had used an appropriate
grading system.
B. Consensus
My impression was that there was massive confusion and no
real conclusions came out except there were many questions about
the ranking scheme. The use of terms made me suspect that the
weighting would cause a skewing of the scale.
Actually the terms used are somewhat vague and need to be
more definitive in toxicological terms.
I felt that we had just begun the discussion and felt very
uneasy about the session.
A-11
-------
Edward Calabrese July 19, 1983
Postmeeting Memoranda
Consensus Topic: Alternatives to the NOEL Derivation Methodology
A. EPA proposes to derive an acceptable exposure level with the use of
regression equations instead of simply dividing NOEL by 100 to
achieve an ADI.
C. Comments:
It is a possible alternative requiring further evaluation.
Future research needs to be directed toward identifying the dis-
tribution of sensitivities in the human population and how this
knowledge could be used in the derivation process. Little work to
date has been published on improving the biological plausibility
of tolerance distribution models (e.g., logit-probit). Yet these
are the very models which in principle are best suited for this
approach. Until EPA is able to offer a reasonable prediction as to
how the variability responses of the test group of animals corresponds
to human variability, this approach will not be based on acceptable
biological grounds. Thus, in principle, this is a good idea because
it tries to stay close to the data. But unless we can relate the
animal model more precisely to the human experience it offers little
improvement over current approaches.
Workshop Topic: High Risk Groups
A. What role should high risk play in the EPA methodology for deriving
acceptable exposure limits?
C. The general need in this area is that of research to identify and
quantify potential high risk groups. Once that is accomplished, it
is crucial to develop better dose-response data which are essential
in the standard derivation process. As in many other cases dis-
A-12
-------
cussed at the workshop, the policy issues are evolving much more
rapidly than the science.
Workship Topic: Ranking Scheme for Dumpsites
A. EPA wants to develop an objective scheme for ranking wastesites.
C. The state of the art is so rudimentary that it is unwise to con-
sider anything said at the workshop as more than a focus for dis-
cussion. The presentation should stimulate further thinking on
the area, and a reassessment of underlying assumptions used.
A-13
-------
July 27, 1983
TO: EPA and Dynamics Corp.
SUBJECT: Post-Meeting Memo
FROM: Tom Clarkson
Multiple Route of Exposure
Issue Can we sum exposures from different routes?
Properties/characteristics of compound to allow multiple route exposures (R. Hartung)
1) The dose-duration patterns should be identical for each route
2) Each route should produce identical systemic effects
3) There should be no effects at the portal of entry
k) There should be no first pass effects
5) The dose-response-time relationships should be known for at least one route
of entry in terms of the actual absorbed dose
Data Base for Route to Route Extrapolation
Listed below in order of completeness are data bases ranging from ideal to virtually
nil:
1) Dose-response-time relationship for all routes of exposure
2) Dose-response-time relationship for less than all routes but with pharmaco-
kinetic data
3) Chronic data for one route but acute data for other routes and pharmacokinetic
data
M Acute data and pharmacokinetic data
5) Acute data only
6) Physiochemical data and structure - activity calculations (e.g. The tnslein
approach)
A-14
-------
With decreasing data bases it is assumed that increasing large safety factors will
be needed in relating route to route extrapolations.
A number of complications were discussed for route to route exposure. First pass
effects and route specific effects, e.g. dioxin, gives both lung and liver tumors by
inhalation but only liver tumors by oral intake.
Dr. Pepelko took extensive notes of the group meeting on Route to Route exposure
and will put together the consensus report. It was agreeed that members of the
group would send their notes to him.
CONSENSUS OPTIONS
I. Intraspecies conversion of dose and duration of exposure for non-cancer effects
It was agreed that corrections for less than life-time exposure would be made
for subchronic and chronic exposure only. The corrections would be based on
the equivalence of fractions of life-span. The corrections would not be made
for acute exposures. Equitoxic dose routes would continue to be calculated
using the cube-root body weight conversion. However, it was agreed that this
subject required further attention. More attention should be given to quanti-
tative physiological parameters available in the extensive literature on com-
parative physiology. A general approach would involve two steps:
STEP 1: Correction for differences in volume of distribution. In general,
this would be on a body weight basis unless there was information
on distribution to specific physical and chemical compartments of
the body. For example, if the agent distributed to the skin, a
surface area conversion would be appropriate.
STEP 2: This should take into account systematic species differences in rates
of elimination of compounds (excretion-metabolism-inactivat ion).
Probably small species such as rodents eliminate chemicals more
rapidly than larger species. The question arises as to how to elim-
inate this systematic difference. It was noted that many functions
related to rates of elimination are approximately at 2/3 power function
of body weight viz. basal metabolic rate, rate of urinary excretion.
It was agreed that STEP 2 permits further research and evaluation.
II. Health Risk Assessment for Less Than Life-Time Exposures
It was agreed that the procedure described by EPA staff was the best available.
For carcinogens the procedure described by Crump (Option 2) was the best
avai table.
A-15
-------
III. APIs based on quoted dose response data
It was agreed that future estimates of ADIs should involve all the dose-response
data and involve a calculation of a "quasi" ADI. This might be set at a 10"1 and
10~2 risk value (Option 2).
IV. Route to Route Conversion
This has already been discussed.
V. Multiple Chemical AH Assessment
Option 2 seemed to be more acceptable than Option 1. However, there are reser-
vations about how the hazard should be calculated. The total hazard index
formula proposed in Option 2 might be appropriate for similar chemicals, have
common mechanisms of toxicity. However, it may not be useful to calculate a
single number representing the total hazard from a mixture of very different
chemicals. In this case, it would be better to look at each individual to see
how it should be weighed in the evaluation of the overall hazard.
A-16
-------
RISK ASSESSMENT FOR MULTIPLE CHEMICAL EXPOSURES
CINCINNATI, OHIO, July 12 - 13, 1983
H. CORNISH
I-V CONSENSUS TOPICS
A. The approach of EPA was to review the data available and previously
discussed and then to see If a consensus could be reached on tech-
niques for evaluation of human risk.
B. Consensus was apparently achieved on many of the approaches but
always with the caveat that more research Is needed and that we do
not have enough Information to do better.
C. The major obstacles to consensus are the degree to which Individuals
are willing to make "best extrapolations" from the data at hand.
Some individuals will never be comfortable with any extrapolation
while others are willing to acccept responsibility for making some
kind of risk estimate on "available data."
D. Comments:
The first morning of the meeting contained a great deal of Informa-
tion from the previous meeting and did not incorporate much new
material. Although the overview was necessary It should have
integrated more of the new material for discussion. Dr. Durkln did
present a new approach to multiple chemical assessment.
Interspecies conversions are often done on the basis of body weight
or surface area. It was suggested that blood levels be emphasized
and utilized wherever possible since this would represent a better
measure of tissue exposure.
Route-to-route conversion evoked considerable discussion. It was
pointed out that the Stoklnger-Woodward model which has been used be
EPA in some instances to convert from inhalation data to oral route
suffers from a number of assumptions. This method, however, was
initially utilized to provide the best possible estimate of a safe
level of exposure for chemicals to which people were being exposed.
Comparable today to dump site exposures. To say, for example that
there Is Insufficient data to make an assessment of hazard at a dump
site may be an abrogation of responsibility and turns the decision
making process back to those who may be the least qualified. I
believe I heard several suggestions that the "Insufficient data"
route be taken. This may be fine for potential exposure situations
but is of no value whee an immediate exposure situation exists.
Perhaps there needs to be some sort of separation of the "immediate
crisis exposure extrapolation" from the "potential hazard exposure
extrapolation." In the latter case it might be legitimate to hold
making a decision until some additional identifiable data were
available.
A-17
-------
VI-VII
WORKSHOP TOPICS
A. Workshop topics were reasonable well stated.
B. I'm not certain that a consensus was reached on these topics but
they evoked considerable discussion and some suggestions for new
approaches.
C. Major obstacles to consensus was lack of information on which one
could make a decision.
D. Comments:
The discussions on structure activity relationships pointed out how
they could be used in risk assessment but there was no presentation
of examples of where it did work and where it didn't work.
Considerable discussion centered on multiple routes of exposure.
Little information is available on which to base predictions of the
contribution of the inhalation route plus the skin route or plus the
oral route as simultaneous exposure routes. Mathematical models may
be useful but research is certainly necessary in this area.
In the "use of experimental data in assessment of health risks"
discussion it was pointed out that the data on hand at a dump site
might Include exposure concentration, duration of exposure, popula-
tion at risk, along with identification of the major contaminants.
To utilize this information to estimate health risks requires Infor-
mation on uptake, species biotransformation, and animal response to
the individual chemical some would say. Hypersusceptible groups may
require greater than normal safety factors. Although a risk estimate
can be done using safety factors, measurement of safety factors in
the exposed population might enhance the reliability of extrapola-
tion. This may be true but I am afraid that for the most part the
total available data base in animals lacks blood level data by which
one could compare expected responses. This points out the need to
encourage the determination of blood levels in animal toxicity
testing.
It seems to me that discussion does not always lead to conclusions
that are usable in risk assessment. Perhaps a future meeting could
be an even more practical workshop. Several approaches could be
used. Give two or three groups the identical chemical and the same
toxicological information and see if they come up with comparable
approximations of a safe exposure level. Another approach would be
to give different groups the same chemical but with varying amounts
of information. For example, can a group that has blood levels In
the exposed population, knowledge of a hypersusceptible group, acute
and chronic animal data really make use of that information and come
up with a different estimate of a safe exposure level than the group
with only some acute data. Perhaps they would come up with similar
risk estimates but with different degrees of confidence In their
A-18
-------
assessment. The latter could often be the case since I'm not sure
ue really know how to make use of much of the data that may say we
must have to make reasonable extrapolations to safe exposure
levels. Actually a safety factor has often been factored in which
expresses the degree of confidence in the data base.
A-19
-------
POSTMEETING MEMORANDUM
To: Dr. Cipriano Cueto
From: Kenny S. Crump J>. C.
Date: June 14, 1983
Subject: June 12-13, 1983 Workshop on Methodology for Risk Assessment
for Multichemical Exposures
Once the general decision is made to adopt the multistage approach
to estimating carcinogenic risks from less than lifetime exposure, there
are a number of sub-issues to be decided—the number of stages and which
stages to be allowed to be affected by the dose. In some cases when
time-to-tumor data are available these can be estimated from the animal
data. Otherwise, some value between 4 and 7, which is typical of human
carcinomas, can be used for the number of stages. This may be the
desired approach even when animal data are available. I concur with
Arnold Kuzmack's suggestion that a table or tables appropriate for use
with typical population distributions would be helpful. It may be that
if the risk is averaged over ages of a typical population, the estimates
will not depend greatly upon the number of stages or the number of
dose-related stages, and, in fact, may give a result which is not very
different from simply averaging the dose over the entire lifespan.
The calculation of a "benchmark dose" (or pseudo-NOEL, or whatever)
appeared to meet with general acceptance, at least for quanta! data.
The risk for which the benchmark is computed needs to be selected. For
moderately large quantal data sets, a risk of about 1% coupled with a
95% confidence limit appears to make the benchmark dose correspond
roughly to the NOEL.
I wish that I had taken time in my presentation to describe and
A-20
-------
Dr. Cipriano Cueto
June 14, 1983
Page 2 of 3 pages
illustrate the benchmark approach to continuous data. The principal
difference is that, instead of being defined in terms of a given
increase in the response probability over that in the background group,
the benchmark is defined as a lower statistical confidence limit for the
dose for which the increase in the mean response over background is a
given percent (say 1%) of the mean response in the unexposed population.
It was pointed out that safety factors applied to benchmarks derived
from quantal data may not be appropriate for those derived from
continuous data. Also, other rationale for safety factors may need to
be put forth for this case. The model used to calculate a benchmark
dose does not appear to be critical so long as the risk value is between
\% and 102. I personally prefer the polynomial regression models.
I liked the suggestion by Rolf Hartung that animal-to-human
extrapolation factors might be determined empirically for
non-carcinogens. Such a research project could shed some useful
insights into the cross-species extrapolation problem. However, I think
there is too much variability in the data for this to be done for single
effects for single chemicals. It could be done for all data on a single
effect, perhaps broken down into classes of chemicals.
The first step would be to collect the data. Perhaps the data base
being assembled by ECAO could be used. Next a well-defined measure of
potency would be estimated for each data set. This measure should
relate to the response in the experimental range and have minimal
statistical variability. The LD^g or the LD^q are potential
candidates for acute effects. Next, an exploratory analysis would be
conducted that relates the potency estimates for different species to a
power of the body weight or some other physiologic parameter. This
analysis could be used to determine if a correlation exists and, if so,
what is the best method for cross-species extrapolation. The "best
method" so determined would have a defensible scientific rationale,
A-21
-------
Dr. Cipriano Cueto
June 14, 1983
Page 3 of 3 pages
since it would be determined empirically from actual data. It probably
2
would not coincide with mg/kg/day, mg/m /day, or any of the other
methods considered by ECAO. Given sufficient data, the method could be
made health-effect specific and also specific to different classes of
chemicals.
KSC:skm
cc: Dr. Jerry Stara
A-22
-------
MEMORANDUM
TO: Cipnan Cueto, Dynamac Corporation
FROM: Patrick Durkin, Syracuse Research Corporation ¦nr
RE: Postmeeting Memo, July 12-13, 1983 Methodology Meeting for ECAO/Cin.
DATE: July 18, 1983
I. Interspecies Conversion of Dose and Duration of Exposure
A. ECAO has used the mg/unit surface area approximation of carcinogens
and up until recently used only the uncertainty factor approach for
non-carcinogens.
B. I perceived a consensus to accept the surface area approach for all
compounds as a default position. Rolf Hartung's suggestion to let the
available data set the exponent seemed to be generally accepted but
examples are needed.
C. No major obstacles.
D. The papers referred to by Hattis and Clarkson deserve careful review.
I will be doing this as part of another task for ECAO/Cin.
II. Less Than Lifetime Exposure for Carcinogens and Toxicants
A. ECAO has used the Druckery approach for carcinogens and has not
formalized the approach for non-carcinogens.
B. The carcinogens led to a clear consensus to accept the Crump proposed
except where clearly contradicted by available data. In such
instances, ad hoc models based on the data should be used. For non-
carcinogens, the approach recommended by ECAO had general acceptance
but needs to be more formally articulated. I personally favor the
blocking over the regression approach but either seems workable.
C. No major obstacles.
D. None.
III. ADI's Based on Quantal, Continuous, or Graded Data
A. EPA has not yet formally articulated a policy.
B. A universal consensus was not reached. I feel that it could have been
reached if more time had been spent. I regard this more as a new topic
than a consensus issue.
A-23
Syracuse Research Corporation Merrill Lane Syracuse, New York 13210-4080 Telephone!315) 425-5100
-------
C. I don't think that any major technical issues need to be resolved to
arrive at a reasonable default position. The major problem was that
several of the participants did not have time to grasp the points which
Dourson and Crump made.
D. Since this issue is somewhat complex, the lack of consensus is under-
standable. I think that if Dourson or Crump were to write a short
position paper and send it to the reviewers, a positive consensus would
be reached.
IV. Route-to-Route Conversion-Pharmacokinetic Approach
A. Up to now, ECAO has used the Stokinger and Woodward approach as the
default position.
B. This approach was generally accepted but with sensible restrictions
(i.e., long t1/2, no first pass effect, etc.)
C. No major obstacles to consensus.
D. No other comments.
V. Multiple Chemical Assessment
A. ECAO has not had a formally accepted approach.
B. For carcinogens, there did not seem to be any serious problem with
using response addition with r = o. For toxicants, 1 was surprised by
the problem that a minority of reviewers had with the dose addition
approach. This conservative approach is the only reasonable way to
avoid presenting results as a vector.
C. What problems arose were at least partially attributable to a lack of
familiarity with the literature.
D. I have no problem with the vector approach - it more clearly reflects
the complexity of the issue. I also have no problem with summing the
vector. It is reasonably conservative and a defensible default
position. Many of the problems could be avoided if EPA adopted a
quantitative dose/response approach for toxicants - something which is
reasonable and which I have long advocated. It would, however, take
some development.
VI. Mathematic Models - Mechanisms of Toxicity
I will not give a detailed response to this presentation because
Dr. 0'Flaherty did not have sufficient time to develop her topic. Her
short discussion of the first pass effect was lucid and instructive but
elementary. Dr. Hertzberg*s introductory remarks set an excellent tone
for the discussion which was not given adequate time.
A-24
-------
VII. Approaches Using Structure-Activity Relationships
A. Dr. Enslein gave a good overview of QSAR approaches but I wish he had
been more specific. He has considerable expertise in this area and I
think that the group would have profited from a more detailed
discussion. Dr. Broderius's discussion of Duluth's QSAR and joint
action work was fascinating but the connection to human risk assess-
ment could have been more clearly indicated.
B. No consensus on the application of QSAR was reached.
C. This entire field requires much more validation before it can play a
major role in human risk assessment.
VIII. Use of Reproductive Effects as Endpoints in Risk Assessment
A. Dr. Manson gave an extraordinarily lucid and instructive presentation.
Given the limited experimental experience of the group in
developmental toxicology, her case study approach was very useful.
B. No consensus was reached on specific or unique approaches for
reproductive effects.
C. I hope that Dr. Manson's participation continues. As we become better
able to utilize her expertise and as she comes to better understand the
goal of and problems In risk assessment, I think that her interaction
with the methodology group could be extremely productive.
NEW TOPICS - These all involve tasks on which I am currently working for ECAO.
I will be preparing very detailed reports which preclude the need
for detailed comments in this memo and I will await reports on the
two workshops which I did not attend.
cc: J. Stara
M. Dourson
C. Beigh
PRD:slh
A-25
-------
POST-MEETING MEMORANDUM
EPA-ECAO 12-13 July 1983, Cmcinatti, Ohio.
1. Hertzberg discussion re model building and testing. I would like to point out
that there is circularity in the procedure of testing an equation with a new data
set. This circularity applies to any model-building effort, and is not limited to
this particular instance.
The probelm is that when one tries to evaluate a model with a new data set
one specifies that the test data must not be outside of the space spanned by
the design data. This is another way of saying that the test set is homogeneous
with the design set. Then, unless there are sampling problems, it is inevitable
that the results obtained with the test set will mimic those obtained from the
design set. Unfortunately, there is no escape from this problem.
2. Route-to-route transformation. This would appear to be a prime opportunity
for the application of structure-activity concepts. What would be needed is data
for the compounds from which the model would be devised for both routes of
administration, for a sufficient number of chemicals. It would seem likely that
such data exist, for example, for rat oral LD-. and inhalation LC5fl. We have
in fact collected such data and believe a model could be constructed. I again
would also like to point out that similar principles exist between any pair of
endpoints for which the data exist, whether thay be obtained by different routes,
in the same or different species, etc.
3. Dermal toxicity. Several times the topic of dermal toxicity arose. I would
like to point out that vast amounts of dermal toxicity data exist in the literatu-
re, and that it would also be practical to develop a model which would relate
dermal endpoints to other endpoints, such a oral and inhalation.
4. Mixtures. After I gave my presentation, it occured to me that the molecular-
orbital calculation methods that Paul Seybold has been using might be applicable
to mixtures. Suppose one considers that a mixture is a jumble ,or juxtaposition
of a number of molecules, then it is conceivable that one could calculate the
interacting forces between these molecules, and estimate the total effect. It pro-
bably would be necessary to take into account the receptor(s) involved, but that
too would seem to be practical, given interactive graphical methods. I will discuss
the possibility with Seybold, and will also keep it in mind at the Gordon Research
conference on the topic , which will be held beginning 25 July.
Kurt Enslein
14 July 1983
A-26
-------
Rolf Hartung, Ph.D.
Consultant in Environmental Toxicology
Professor of Environmental Toxicology
University of Michigan
3125 Fernwood Ave.
Ann Arbor, Michigan 48104
(313) 971-9690
POST-MEETING MEMORANDUM
July 20, 1983
Dr. Cipriano Cueto
Dynanac Corporation
11140 Rockville Pike
Rockville, Maryland 20852
Dear Dr. Cueto:
The EPA Methodology meetings on July 12 - 13 in Cin-
cinnati generated several new ideas which should serve to
improve the risk assessment methodology. However, because
of the size of the meeting, complete consensus was rarely
achieved on any of the issues discussed. If one considers
that many aspects of risk assessments must rely on extrapo-
lations which are based on "reasonable or plausible hypo-
theses" which have not been verified, then such a lack of
consensus is understandable. In my opinion, it is more
valuable to explore the range of alternative hypotheses and
their assumptions, than to seek consensus. I abhor the idea
of establishing consensus by majority vote.
Many of the methodological aspects which were discussed
had been wrestled with previously. I fear that some of the
methods may be accepted just because they have been reiter-
ated often enough, and because there is simultaneously a
lack of data to support or to refute a specific method.
a-27
-------
I would like to discuss specifically several methods which
seem to have those characteristics.
Exposures Multiple Chemicals:
This has been one of the major themes during the recent
methodology meetings. The topic is obviously very relevant
and very important for the evaluation of risks from expo-
sures in the "real world". Discussions of the most plau-
sible extrapolation models for the evaluation of multiple
chemical exposures invariably settle on the various aaditi-
vity models. What is not fully discussed is that all of the
data on additivity which have been found useable to explore
the phenomenon have been gathered for high dose exposures,
and a majority have been derived from single dose exposures.
Almost all of the exposures have been to two chemicals only.
Such combinations of factors are likely associated with non-
linear kinetics. Whether "additivity as a rule" predomi-
nates for chronic low level exposures to multiple chemicals
is completely speculative at this time, and all of the
modelling and arguing in the world will not alter that
status. We simply need to do research to find out what is
happening! By identifying this research need, I do not
propose a proliferation of acute studies which seek as the
main goal the biochemical basis for specific two-chemical
interactions. The basic need is to do chronic multiple
A-28
-------
chemical exposures near previously identified LOAEL doses
using protocols which can be used to test various interac-
tion models.
Low-Dose/Lonq-Term Extrapolations Carcinogens: I
note less and less in-depth discussion concerning the cur-
rently accepted modification of the multi-stage extrapola-
tion model. The discussion of less-than life-time exposures
to carcinogens revolving around early-stage vs. late-stage
carcinogens appears to imply that the coefficients in the
exponential polynomial which constitute the multi-stage
model, have real biological meaning. Every study of various
mathematical dose-response models used for very low risk
extrapolations has demonstrated drastic divergence between
models for very levels of risk. While attempts have been
made to justify the exclusive "scientific validity" of the
multi-stage model, such attempts do not do justice to the
actual scientific uncertainties which are involved. Issues
related to the degree of conservatism in the selection of
extrapolation models incorporate personal and/or political
value judgements which should be considered issues separate
from scientific uncertainty pertaining to accuracy and
precision of the extrapolations.
A-29
-------
&£gcies jyj Spgcies fiXtraFgHataongs This is an area of
continuing controversy and active discussion. That is pro-
bably because this type of extrapolation appears to be
susceptible to better resolution through research by employ-
ing pharmacokinetic and pharmacodynamic techniques, studies
of metabolism and of macro-molecular binding in relation to
measured adverse effects. Various forms of such research
are being actively pursued in many institutions, but many of
the research protocols do not seem to be constructed .with a
consideration of an ultimate end-use related to the species:
species extrapolation problem. The species differences due
to excretion rates and residence times in the body were
specific issues which had not been that thoroughly consi-
dered in previous meetings. These issues require further
elaboration. That can be done more readily by a small task-
group rather than a large committee. At this time it is
clear that a simple cube root conversion may be useful as a
default condition, but there seem to be many conditions when
that conversion does not apply. The influence of metabolic
activation vs. inactivation in relation to the cube root
conversion was brought up again, but once more there was
insufficient time and effort focussed on that issue. In the
overall scheme of things these species:species extrapola-
tions should take the place of the arbitrary 10X safety
A-30
-------
factor traditionally employed for species differences in
setting A.D.I, levels.
Multiple-Route Exposures: the evaluation of multiple route
exposures was analyzed as a special case of routerroute
extrapolation, and in the case when exposures by different
routes produced different effects as a special case of
multi-chemical exposure. The simplistic additive approach
suggested earlier, and as summarized in the pre-meeting
information, was considered to be inadequate.
The approach taken to the problem was to consider first
those conditions under which a multi-route exposure evalua-
tion could be made with confidence. In order to keep the
problem tractable, the problem analysis was considered for a
single species only. An evaluation with confidence can be
made ii and only il all of the following conditions are met:
1.- the dose-duration of exposure patterns are
identical for all routes of exposure;
2.- the systemic effects are identical for all routes
of exposure;
3.- no route-specific effects are found at the portal
of entry;
4.- there are no first pass effects;
5.- the dose-duration of exposure-response surface is
known in terms of absorbed dose (or: the relationships
A-31
-------
between applied dose and absorbed dose are known for each
route).
Points 1 and 4 are likely to become unimportant if the
chemical is cleared very slowly from the body.
If all of these conditions are met, then the total
absorbed dose could be estimated, and subsequently the
systemic response in terms of absorbed dose could be
evaluated.
It was recognized that these preconditions would only
rarely, if ever, be met for a specific compound. It was
also pointed out, that as these various pre-conditions are
violated, then the ability to predict the consequences of
exposures by various routes deteriorates in ways that are
not easily predictable.
If there were route-specific effects, then a simple
summation of absorbed doses would probably be misleading,
since interactions between effects would not be taken into
account appropriately. Under such conditions (if dose/
response relationships were known for each route of expo-
sure), the problem could be treated as a special case of the
simple joint action additivity model as postulated by
Bliss(1939). Such a model could readily be formulated in
terms of external dose, rather than absorbed dose.
Since exposures by multiple routes appear to be common,
and since dose/response information is often imperfectly
A-32
-------
knowr. f-r ;he various routes of exposure, it is tempting to
attempt estimations of the effects of multi-route exposures
on the basis of mixtures of chronic and acute data or on the
basis of chronic data alone. Such estimates may be very
unreliable and misleading.
Overall, I judged the meeting to be productive, flow-
ever, a number of the issues which were raised in the
meeting could not be adequately considered because of time
constraints and because of the number of participants invol-
ved. Such issues should be handled is smaller groups with
smaller agendas.
I hope that you find this information useful.
Rolf Hartung
A-33
-------
MIT E40-239
(617) 253-6468
July 23, 1983
Cipriano Cueto, Ph.D., Program Manager
Dynamac Corporation
Enviro Control Division
The Dynamac BJfuilding
11 HO Rock vi lie Pike
Rockvilie, Maryland 20852
Dear Dr. Cueto:
I very much enjoyed the opportunity to participate in the recent EPA
risk assessment meeting on July 12-13. Please feel free to call on me
again if 1 can be of help in the future. Enclosed are my post-meeting
memorandum, expense form, and receipts.
Sincerely,
¦ V ' ' 1 ' ¦ !
v !l m .. f - I /
Dale Hattis,' Ph.D.
A-34
-------
July 23, 1983
To: Cipriano Cueto, Ph.D.
From: Dale Hattis, Ph.D.
Re: Postmeeting Memorandum for Workshop on Risk Assessment Methodology
I. Interspecies Conversion of Dose and Duration of Exposure for
Non-Cancer Effects
The most common approaches at present in risk analyses by EPA and
others are options number 1 and 2, as outlined in the pre-meeting briefing
paper--simple mg/kg body weight or mg/body surface area conversions. The
major EPA proposals were to modify these based on either
(option 4) pharmacokinetics of the chemical, including activation or
detoxification, and
(option 5) goodness-of-fit to the toxicity data
Both of these have their merits. If sufficient cross-species toxicity
data is available, option 5 is clearly to be preferred, as it would
implicitly account for both pharmacokinetic differences and any
differences related to the receptor(s) for toxic effects. If good
cross-species toxicity data 1s not available, but some cross-species
pharmacokinetic information can be obtained, then option 4 should
generally provide the best guidance. In both cases, however, the choices
should not be limited to simple mg/kg or mg/surface area rules, but should
allow for a gradation of body weight" scaling factors, as indicated by
the available information (see further discussion below).
Option 3 (mg/kg brain weight for CNS effects or, by extension
mg/target organ weight for other effects) may have been constructed on the
premise that one should expect effects in a target organ to be similar
across species for doses of chemical that produce similar concentrations
of active chemical in the target organ. This is a reasonable premise.
However, the concentration of active chemical in the target organ will
generally depend on the total body distribution and metabolism of the
chemical, not simply the volume or weight of the target organ. Therefore
option 3 is unlikely to Improve the accuracy of dose rate projections
except in the rare case where the great bulk of the chemical 1s
distributed to the brain (or other target organ).
Most of the discussion 1n the workshop centered on what general
rule(s) should be adopted in the usual case where one does not have good
cross-species chemical-specific pharmacokinetic or toxicity information.
Based on experience with biological half lives of methyl mercury and
various drugs, there appeared to be general agreement that mg/surface area
dose conversion would more often be a more accurate predictor of internal
body (dose X time) of active chemical than simple mg/kg body weight
conversion.
There was considerable interest, however, in a paper I brought to the
attention of the group (Boxenbaum, H.: "Interspecies scaling, allometry,
physiological time, and the ground plan of pharmacokinetics." J.
Pharmacokinetics and Biopharmaceutics 10 201-227, 1982). A major thrust
A-35
-------
-2-
of this paper is that basic metabolic rates in organisms scale to (body
weight)*'5. Therefore the turnover time of the average body constituent
is proportional to
(body weight)V(body weight)*^ = (body weight)*^
Boxenbaum assembled data which indicated that in general the interspecies
scaling of biological half lives of drugs tended to cluster around this
same value, although the caveats must be attached that
o there were considerable departures for individual drugs (ranging
from body weight exponents of .06 to .428 for the eight examples
listed), and
o in three of the eight cases, the data for humans were considered
an outlier from the general pattern and were excluded from the
analysis
I think that useful insights might be obtained by assembling more of this
kind of information, and attempting to discern patterns in (1) which kinds
of chemicals scale with larger vs. smaller body weight exponents, and (2)
which kinds of chemicals appear to give rise to anomalous human data; in
which direction is the data anomalous (toward more persistence/greater
risk or toward less persistence/lesser risk) and how large 1s it? In the
absense of other information, I think that mg/kg doses of chemicals should
probably be scaled by a (body weight)»25 factor rather than the (body
weight)'/3 which corresponds to the surface area conversion or the (body
weightjO which results from the simple mg/kg body weight procedure.
there was not much discussion of the propriety of
fraction-of-1ifespan scaling of duration of dosage for non-cancer effects.
It is difficult to entirely separate consideration of dose duration from
the dose rate problem discussed above. In the absense of other
information, lifespan scaling seems reasonable. However, I would think
one could place the assumption on firmer footing if at least a limited
number of comparisons could be made of the time X dose-rate tradeoff for a
few defined toxic endpoints (e.g. LD50, x% impairment of some neurological
function) for a few species with differing lifespans. If the dosing-time
increase needed to halve the required dose to produce several different
endpoints really were directly proportional to lifespan in the model
systems, this would support use of the proposed rule.
II. Health Risk Assessment for Less than Lifetime Exposure
Toxicants
The two options outlined in the briefing package do not differ much.
Essentially, option 1 calls for a grouping of the data into four discrete
time-lengths of exposure, and option 2 contemplates use of ungrouped data
to estimate a line for any duration. Given a choice, I would think that
the procedure using ungrouped data is slightly better in that it uses all
the available information in a systematic and non-arbitrary way (grouping
tends to sacrifice some information).
A-36
-------
-3-
I do not recall very much explicit discussion of the differences
between these options by participants in the workshop. I remember that I
expressed some reservations about the adequacy of the 10-fold safety
factor for within-species and surface-area-converted data (see discussion
of hypersusceptibility issue below) and others may have shared this
concern.
Carcinogens
The presentation by Crump of the results of his calculations under
the multistage model (option 3) was quite cogent, although some present
may have had some difficulty grappling with the technical details of the
modeling. Crump's approach has clear theoretical superiority to the
present procedure (option 1) of ignoring the expected age differences in
cancer dose-effect relationships. There is, of course, the considerable
practical difficulty of deciding whether one is dealing with an
early-stage or a late-stage carcinogen in specific cases, and deciding
exactly how many stages to assume. These uncertainties, however, can be
explored with sensitivity analyses and presented as part of the overall
uncertainty of quantitative carcinogenesis risk projections. I heard no
serious explanation or advocacy of option 2, and I think it 1s preferable
to base projections of risk on models (like Crump's) that are firmly
rooted in some sort of hypothesized mechanisms for the effects being
studied.
III. API's Based on Quanta!, Continuous or Graded Data
The options here were basically
1. Fit a mathematical model to the animal dose response data as 1n
carcinogenesis risk assessment and Interpolate all the way down to
risk levels of concern (10"® or 10"? or so—I think the 10'5
suggested in the briefing package is quite high for an acceptable
risk level except under unusual circumstances with very clear
benefits for risk assumption.)
2. Fit the same kind of models but project down only to some
"benchmark" risk level (10-' or so). Beyond that, use "safety
factors" to establish ADI's. Safety factors could differ for
different kinds of effects depending on the severity of the effect,
the quality of the data, and the mechanism of the effect.
I have misgivings about both these approaches, but with current
information and some modifications, option 2 is preferable. The fatal
difficulty for option 1 is that the steepness of the dose response
relationship observed for a classic toxic effect (one that is
characterized by individual thresholds for response) depends entirely on
the diversity of the exposed population. For example, if the exposed
population is relatively uniform, the dose that produces the response in
80% of the exposed individuals might be only 1.5 times the dose that
produces the response in 20% of the exposed individuals--making for a very
A-37
-------
-4-
steep dose-response curve. Any mathematical model applied to such dose
response data and used to directly project human risks at all dose levels
would necessarily incorporate an expectation that this steep kind of
relationship would also be seen in humans. One thing that we can be very
confident of, however, is that a typical human population exposed to the
same agent will be very much more diverse in individual response
thresholds than the inbred and age-restricted populations of experimental
animals that are used for testing. In general, therefore we must expect
that the population dose-response relationship in people would be much
shallower—and carry relatively greater risks at low doses--than would be
expected if people were as uniform in their response thresholds as the
test animals. There is just no way to get appropriate information about
the distribution of human response thresholds directly from the animal
data that will be available in most cases.
The (option 2) approach of modeling down to a selected (relatively
high) risk level and applying a "safety" factor 1s objectionable in that
it mixes a finite/quantitative risk assessment approach with the more
traditional tool of determining a dose thought to pose no risk by applying
some standard rule of thumb beyond the level where data 1s available. In
view of the difficulties of full quantitative risk assessment alluded to
above, however, it is probably the best that can be done for most
chemicals at present. However, with the majority of the people in the
hypersusceptibility work group, I would suggest that different "safety"
factors should be used for different kinds of effects, depending on
o The likely bredth of the distribution of susceptibility 1n the
exposed human population for different chemicals and effects (I
presented a detailed outline for work in this area during the
hypersusceptibility study session)
o Policy judgements about the severity and socially desirable
protective posture for different kinds of effects.
In this context, as a technical matter I don't like the use of the terms
"safety" or "no-effect level" because these imply population thresholds
and zero risk. As Dr. Crump pointed out, his modeling approach to compute
a benchmark exposure level from the animal data that gives rise to a
defined proportion of response in the target population should not be
portrayed as calculating a level at which one expects no (zero) animals to
show the response. The "no effect" levels ususally determined depend
simply on the size of the population of animals studied and the bredth of
the distribution of individual thresholds in that population. Similarly
if the computed benchmark levels are used to calculate ADI's using
"safety" factors there is an unfortunate implication that the calculated
ADI represents a level of exposure which should produce absolutely no risk
in the human population. In fact, however, if the chemical worsens some
physiological function that in some people is already below levels
necessary for normal functioning (e.g. oxygen transport to the heart
muscle in people with agnina pectoris) then in prinicple there can be no
threshold of exposure below which no people will be adversely affected.
A-38
-------
-5-
IV. Route to Route Conversion
The presentations in this area at the meeting were excellent and, as
I remember, evoked no serious dissent. The proposed consensus approaches
in areas B {conditons acceptable for extrapolation) and C (extrapolation
procedures) in the pre-meeting briefing package also seem quite reasonable
as stated. I would add only a modest qualification to A (exposure routes)
to the effect that while dermal absorption is indeed negligible as a route
of exposure in most general environmental contamination situations, it
sometimes is important when evaluating exposures from chemical products
within the province of the Toxic Substances Control Act and the Consumer
Product Safety Act (e.g. lubricating oil additives, household window/wall
cleansers, dyes used in inks, chemicals used in photographic development
and copying machines). People in the Office of Toxic Substances have
considered dermal absorption important enough to fund the compilation of a
data base on dermal absorption kinetics and toxic effects from chemicalIs
administered dermally within the SPHERE component of CIS.
V. Multiple Chemical Assessments
The options here are either to essentially ignore the presense of
multiple chemicals (choosing only the single chemical of greatest
concern--option 1) or to construct some kind of index designed to be
proportional to the overall level of concern one might have if the risks
posed by all the chemicals present simply summed up to a total risk
(option 2). Although there was vigorous dispute in the meeting as to the
propriety of a simple additive index of concern crossing diverse kinds of
chemicals and toxic effects, I think this is an eminently reasonable
quick-and-dirty approach for general screening/priority setting purposes.
It may not be the way one would go about a detailed full-dress
quantitative assessment of the risks of an exposed population if one
really wanted to know how many people are likely to be hurt to the
greatest accuracy possible with available information. However, if the
need is to approximately rank for attention some large number of waste
sites witti many chemicals per site and huge uncertainties as to the nature
and extent of exposures, no better than a quick-and-dirty approach is
feasible or reasonable within the constraints of time and information
available to make real choices.
VI. Hypersusceptibility Workshop
There was substantial agreement by the participants in this workshop
that getting some kind of quantitative handle on variation in
susceptibility 1n the human population was an important area of research
for improvement of current EPA risk assessment practices. In general it
was felt that this kind of consideration should be built into the
calculation of ADI's, rather than (as suggested in pre-meeting material)
incorporated as an add-on to estimates of populations at risk. There
seemed to be good agreement that in calculating ADI's different safety
factors should be used for different kinds of adverse effects depending on
(1) the likely population distribution of sensitivity for the effect in
A-39
-------
-6-
exposed humans and (2) policy judgements as to the degree of protection
desirable (i.e. "acceptable" risk levels) for specific effects.
As part of the workshop, I outlined an approach for research and
modeling to define distributions of susceptibility for "classic" toxic
effects (those with thresholds). Enclosed is a copy of this outline.
Some in the group felt it would be more important to define distributions
of susceptibility for kinds of effects without thresholds (e.g. cancer and
also "chronic cumulative" effects by my taxonomy). I think the issue is
important in the latter cases, but is the main determinant of quantitative
human risk for the threshold-type effects. Prior to the meeting, I
provided to Dr. Erdreich a copy of a proposal I submitted some time ago to
NIOSH to compile information on the human variability of parameters likely
to produce differences in susceptibility to carcinogens. I would welcome
an opportunity to do further work on this subject for all kinds of adverse
effects.
VII. Exposure Workshop
The main item of consensus to come out of this meeting was that there
should probably be more interaction between the groups estimating
exposures and the groups calculating dose response relationships and
numbers of people affected in normal EPA risk assessments. Without
interaction, there is a substantial chance that the the exposure group
will not provide information to the risk assessors integrated in
appropriate time intervals (in the light of the expected dynamics of
production of the adverse effects) or subdivided by the population groups
that may differ systematically in dose-response. There is also some
chance that the risk assessors could simplify the task of the exposure
estimators by advising on the chemicals (or "marker" proxy exposures) of
greatest interest for measurements and modeling.
A-40
-------
Dale Hattls*
7/13/83
Comments on the Treatment of "Hypersusceptibility"
in Risk Analyses for Non-Carcinogens
1. Subject should be considered central to the projection of risk for
types of effects that occur when a stimulus overwhelms some specific body
compensatory processes (I.e. not information-change type mechanism or
accumulation of small irreversible damage steps).
—framework of traditional toxicology: problt analysis presumes log
normal distribution of thresholds
--importance in past rulemakings
o carbon monoxide/angina sufferers
o lead/children with pica/children at risk for mental
impairment
o asthma sufferers/sulfur oxide standard
2. Comments on the proposed approach for considering
"hypersusceptibiHty" 1n agency risk assessments.
—the proposed approach assumes a discontinuous-population-subgroup
model of "hypersusceptibility"; however much of the variability
between individuals in susceptibility may appear as continuous due to
a host of factors acting as quantitative characters, and not based on
obviously discemable membership of people in discrete, countable
subgroups.
—the proposed approach appears to implicitly assume that the average
"hypersusceptlbl e" person 1s only about twice a susceptible as the
average non-"hypersusceptible". (If the difference were really to be
as small as this, it wouldn't be worth much bother amid the other
uncertainties of quantitative risk assessment.)
—the propsed restriction to "typersusceptlble" groups comprising at
least IS of the population, while it flows naturally from the propsed
calculation method, 1s arbitrary and potentially seriously
misleading. Consider, for example, a case where there exists a .1%
subgroup that 1s ten thousand times as susceptible to a toxin as the
average individual. Quantifying the size of the population that Is
very much more susceptible than average 1s the essence of a
quantitative risk assessment for a threshold character with widely
varying sensitivity.
~M.I.T. Center for Policy Alternatives
A-41
-------
-2-
3. Possible analysis of different sources of variation in susceptibility
--exposure is effectively more for the subgroup because of either
biology or behavior
o children probably eat more dirt than adults, and children
with pica eat particularly more
o children tend to play on old fields more than adults, some of
which used to be toxic waste dumps
o children simply breathe and eat more per body weight than
adults, even discounting behavioral differences that relate to
what they eat
NOTE: THESE TYPES OF DIFFERENCES ARE NOT GREATLY DIFFERENT FOR
DIFFERENT KINDS OF CHEMICALS—DATA OBTAINABLE BY POPULATION
INTERVIEW/MEASUREMENT SURVEYS FOR CHEMICAL-EXPOSING BEHAVIORS
—parameters determining normal /abnormal function are closer to (or
even beyond) the boundary of abnormality for the subgroup. Two
kinds:
o disease-based—angina/oxygen deprivation (even more extreme
case is people in the process of having a myocardial infarction)
o other interacting exposure—smokers/prior CO exposure ties up
hemoglobin that would otherwise be available, thus probably
making them more susceptible to the effects of aromatic amines
causing methemoglobinemia—not because the aromatic amine ties
up any more hemoglobin, but because they have less "reserve
capacity" to begin with
INFORMATION ON THESE TYPES OF VARIATIONS MAY OFTEN BE OBTAINABLE
FROM CROSS-SECTIONAL IHTERIEW SURVEYS OF THE TARGET POPULATION
FOR THE PREVALENCE OF RELEVANT CONDITIONS AND POTENTIALLY
INTERACTING EXPOSURES. THE RELEVANT CONDITIONS, HOWEVER, WILL
VARY FROM AFFECTED-PARAMETER TO AFFECTED-PARAMETER THUS ANALYSIS
OF THIS KIND OF VARIATION MUST BE SPECIFIC TO THE PARAMETER,
THOUGH NOT NECESSARILY TO THE CHEMICAL. DIFFICULTY: IN MANY
CASES THE PRECISE BIXHEMICAL DISTURBANCE TO HOMEOSTASIS WILL
NOT HAVE BEEN DEFINED FOR SPECIFIC CHEMICALS CAUSING CHRONIC
TOXIC EFFECTS.
--there is more parameter change/unit dose of chemical for the
subgroup, because of biological differences 1n
o pharmacokinetic factors
the absorption of the chemical (fraction of the amount
taken into the lungs or the GI tract that finds its way to
the blood stream)
a-42
-------
-3-
the persistence of the chemical or its active metabolites
in the body (half-life or more complex parameters of
decline with time)
o the fraction of the chemical that is metabolized via
active/dangerous vs. inactive/safe metabolic pathways
o factors related to the receptor for toxic action
affinity of receptors for the toxic agent, vs the affinity
for natural competitors (e.g. the affinity of
acetylcholinesterase for cholinesterase inhibitor toxins,
vs affinity for the natural substrate acetylcholine)
number of receptors, in relation to the minimal number
required for normal function (I.e. the functional reserve
at the micro level)
INFORMATION ON THESE FACTORS WILL NOT GENERALLY BE OBTAINABLE
FROM CROSS-SECTIONAL INTERVIEW SURVEYS. ANALYSES MUST BE BASED
ON EITHER
—CLINICAL OBSERVATIONS OF THE RESULTS OF MODEL EXPOSURES
TO THE CHEMICAL UNDER STUDY OR PHARMACOLOGICAL ANALOGUES,
—OR (WITH CONSIDERABLE DIFFICULTY) PROJECTIONS FROM ANIMAL
STUDIES.
Suggestion: Divide susceptibility/variability analyses up Into these
three components.
—Have a reasonably standardized analysis of susceptibility for the
exposure-related factors, depending only on the routes of anticipated
exposure to the target population.
--Do a series of standard analyses of population variability for
different well-characterized physiological functions that are
affected by toxic agents (e.g. oxygen transport, nerve conduction).
For substances that affect unknown physiological parameters, use a
"typical" variability profile borrowed from one of the
well-characterized functions for "best estimate" calculations, and
some more extreme assumption of variability for "conservative" or
"worst case" projections.
—To the extent possible, analyse pharmacokinetic and
receptor-related variability chemical-by-chemical. Where
chemical-specific data is not available, 1t may often be necessary to
adopt some general rules based on previous experience with specific
chemical classes or chemicals affecting specific receptors.
Based on this three-tiered analysis, construct a stochastic simulation
model (combining randomly-selected values for each of the relevant
susceptibility factors within the three tiers) to compute the overall
distribution of population sensitivity for the chemical.
A-43
-------
THE UNIVERSITY OF TEXAS MEDICAL BRANCH
GALVESTON, TEXAS 77550
Department of Preventive Medicine and Community Health
Division of Environmental Toxicology
Phone:409-761-1803
July 15, 1983
Cipriano Cueto, Ph.D.
Project Director
Dynamac Corporation
The Dynamac Bldg.
11140 Rockville Pike
Rockville, MD 20852
Dear Dr. Cueto:
Re.: EPA's Workshop on Methodology for Risk Assessment for Multi-
chemical Exposures held July 12-13, 1983, Cincinnati, Ohio
I was impressed by the scientific expertise represented in this
meeting. I felt that the topics were selected with a great deal of care
and the level of presentations were excellent. My basic criticism is
that our discussion time was so severely limited. This was especially
so in our individual workshop sessions. In the session on ranking I
was particularly discouraged because a single presentation accounted
for almost all of the allotted time. I suspect try adverse reaction was
on a very personal level. I had prepared some slides that I hoped would
lead to a more productive way of ranking toxic chemicals. Perhaps in
a future meeting with a similar number of participants we would be more
productive if we were split into smaller groups Initially to discuss
specific topics.
I will put some of my notes together and send you a more definitive
critique at a later date.
Yours truly,
Marvin S. Legator, Ph.D.
Professor and Director
MSL/tml
A-44
-------
Postmeeting Memorandum of Jeanne Manson
Session III: B, A2: Ranking the Severity of Effects.
Presentor: P. Durkin
RE: Table 1. Rating Values for Toxic Effects
I am somewhat concerned about the rating score for point #8, which includes the
statement "Any decrease in reproductive capacity, any evidence of fetotoxicity". My
concern is that these events could occur as secondary or tertiary effects of generalized
toxicity, and may not represent the irreversible or significant toxicity implied by the
high rating score. In the case of reproductive capacity, transient decreases in sperm
count or dysmenorrhea can occur as a result of general systemic toxicity, given that the
function of the reproductive system is dependent upon the highly integrated function of a
number of other systems. Consequently, I think a distinction should be made between
reversible decreases in reproductive capacity versus long-term decreases in reproductive
capacity and decreases in reproductive capacity that occur with general systemic
toxicity from those occurring in the absence of other types of toxicity. Reversible
decreases in reproductive capacity, and/or decreases in reproductive capacity
accompanied by general systemic toxicity could be placed under point 7 to reflect
detectable decrements in organ function.
A clearer case can be made for underrating "any evidence of fetotoxicity" given
the often-quoted rule of Karnofsky's law, which is that any agent can be shown to be
teratogenic (fetotoxic) if administered at high enough levels at the appropriate time of
gestation. I think if "fetotoxicity" were more specifically defined, then a lot of
confusion could be avoided. For example, decreases in fetal body weight and increases in
delayed ossification accompanied by maternal toxicity could be placed under point 4.
Excessive embryolethality accompanied by maternal toxicity could be placed under
A-45
-------
point 7. Any teratogenic effect with maternal toxicity could be listed under #8. The
most important point, however, is that to leave the statement "any evidence of
fetotoxicity" without qualifying whether or not the fetotoxicity occurs at maternally
toxic doses is the same as incriminating all compounds tested at maternally toxic doses.
A-46
-------
M. a. MEHLMAN
7 BOUVANT DRIVE
PRINCETON, N. J. 08540
July 25, 1983
Dr. C. Cueto
Dynamac Corporation
Enviro Control Division
The Dynamac Building
11140 Rockville Pike
Rockeville, MD 20852
Dear Dr. Cueto:
I would like to comment on the EPA's Workshop on Methodology for Risk
Assessment for Multiple Exposures which was held on July 12-13, 1983 in
Cincinnati, Ohio.
After a somewhat slow start on July 12, due to prolonged discussion of
issues, the Workshop picked up speed on the second day. It was clear that
in order to deal with these complex topics, it will be necessary to
commission one or two people to develop position papers. Most of the
groups that meet in separate sessions have recommended so.
The session that I chaired, The Use of Exposure Data in Health RisK
Assessment, has raised concern on the following:
• We assume that the type and amount of wastes will be reasonably well
characterized (but may be deficient in some cases).
• Exposure assessments are resource-1imi ted and data-limited, and
exposures are variable in space and time; therefore, exposure cannot
be characterized by a single number.
The following data are desirable for characterizing exposure:
• Monitoring data from air, water (surface and ground) foodstuffs
(especially fish) and soil must be determined to adequately establish
the level of substances present.
• Tissue/excreta monitoring in humans or animals from the area is
extremenly valuable in assessing exposure.
• Human behavioral observations in the area should be made to identify
exposure routes or sensititve individuals.
• Attention should be given to proper locations of monitoring wells.
A-47
-------
Dr. C. Cueto
July 25, 1983
2
• The output should be explicit as to the variability associated with
exposure, both in space and time.
• Monitoring data and exposure assessements come from a wide variaty
of sources within and without EPA. Close coumunication between those
responsible for each phase must be encouraged, especially if a second
phase of monitoring can be done.
• We recommend that the Agency commission a paper discussing integration
of environmental monitoring, exposure, and health data, including the
role of epidemiology of exposure.
This Workshop has accomplished it's goals by thoroughly discussing the state
of science of topics that were assigned and by identifying and recommending
steps to be taken to deal with these topics.
Sincerely,
M. A. Mehlman
A-48
-------
Post meeting coments, W. J. Nicholson
A. Interspecies of conversion of dose.
I would suggest that our understanding of interspecies conversion of dose,
at least for some end points such as cancer, is insufficient to allow a
specific policy to be established. It would appear that the factor to be
utilized depends strongly upon the metabolic processes involved the toxic
effect in question. As an example, I would cite the emerging data on the
carcinogenicity of vinyl chloride in humans as compared to that in rats.
Data from the extensive series of Maltoni suggest that 15% of rats ex-
posed to an air concentration of 250 parts/million (8 hours/day, 5 days/
week, 52 weeks - if3 of a lifespan) develop hemangiosarcoma of the liver,
(note an adjustment to an 8 hr. day)
Data from human epidemiology (Nicholson Proc. Int. Symposium on hazards
related to plastics and synthetic elastomers. Helsinki, Nov. 1-°-?). sug-
gested that from 200 to 600 hemangiosarcomas of the liver will eventually
occur among U.S. VC-exposed workers from exposures that occurred prior to
1975. These cancers will occur among approximately 5000 workers exposed
in the industry to 200-500ppm for periods from 10 to 40 years (approxi-
mately 5-/7 of a lifetime). Thus the equivalent human risk is 4% to 12%
over the remaining lifespan of the workers.
One can assume that man inhales 40 times more VC (Scaling as 2/3 - oxygen
2/3
consumption scales as W .If the risk of HSA in the liver is propor-
tional to the VC concentration, *
less than the rat (if all VC inhaled is metabolized). This suggests
that for the same air concentration, the time concentration will be
_ ----- „ reasonable agreement consider-
ing all the uncertainties.
man. The observed risk in
A-49
-------
2
The above, however, completely neglected the species dependence of acti-
vation. Gehring et all have explained the observed non-linear dose re-
sponse function for Maltoni's experiments in terms of Michaelis-Menten
kinetics in which the maximum rate of conversion of vinyl chloride to the
reactive epoxide is 5706 migrograms/4 hours. It is not know whether
similar Michaelis-Menten constants will apply in man. If they are signi-
ficantly different, the concentration of ractive metabolite in the liver
will differ from the naive estimate above.
If the kinetics (constants in Michaelis-Menten equation) of metabolic
activation are species dependent, significantly different species conver-
sion factors would obtain depending on whether a carcinogen is direct
acting (and metabolism reactivates) or requires activiation.
B. Procedure for 'estimating risk for fraction of a lifetime exposure
The multistage model proposed by Crump and Howe would appear to be
eminently suitable for estimating lifetime-risk for less than lifetime
exposure for most carcinogens. However, the mode of action of a carcino-
gen must be carefully considered as different less than lifetime risks
would be estimated depending upon whether material is acting predomin-
antly as an initiator, (early stage carcinogen) or a promotor (late stage
carcinogen). Thus, as with interspecies conversion, knowledge of the
mechanism of action, material is important in the application of a model.
The procedure for estimating a risk for less than lifetime exposure for
non-carcinogenic effects can be established easily if the procedure dis-
cussed for calculating a pseudo NOEL or 1% NOEL become generally utilized.
The less than lifetime dose can be quantitatively established by consider-
ing the calculated 1% NOELs established at different fractions of a life-
A-50
-------
time. A regression line utilizing these values can be calculated with
each value being weighted by the reciprocal of the calculated variance
of the 1% NOEL level. This procedure can of course work only if there
are sufficient data available in each discrete fraction of a lifespan
period.
API's based on quantal, continuous or graded data
Whenever possible, continuous data should be utilized for estimates of
biological effects rather than quantal data. For example, the average
weight loss of a group of animals to a given concentration give a far
more accurate measure of effect than the percentage of animals that are
below some specified "normal" level. Such continuous data could be
incorporated into a model (Probit, linear, etc.) in which a 1% effect
is calculated. With such a calculated 12 effect, an uncertainty factor
could be established for regulatory purposes.
Safety factors
Much more data are required for the establishement of safety factors.
Currently, a factor of 10 is utilized to account for variability within
species (man). This may not be at all adquate considering the special
sensltifity of children, often coupled with greater exposure to toxic
environmental agents.
One useful set of data on intraspecies variability would be to look at the
14-
variability of biological sensitl^ to some agent in a purebred strain
of animals and in a variety of inbred strains of the same animal. It
would then be possible to calculate the difference in doses for an effect
at the 0.00 1% level that would be estimated from measured effects at much
higher doses.
A-51
-------
©
Clement Associates, Inc.
1315 WILSON BOULEVARD SUITE 700 ARLINGTON VA 22209 <703) 276-7700
August 25, 1983
MEMORANDUM
TO: Jerry Stara, ECAO, USEPA, Cincinnati, Ohio
FROM: Ian Nisbet, Clement Associates, Inc., Arlington,
Virginia
THROUGH: Christopher Dippel, Dynamac Corporation, Rockville,
Maryland
RE: Post-Meeting Memorandum, Workshop of July 12-13,
1983
General Comments
I found the workshop helpful and stimulating, and I appre-
ciated the opportunity to take part in it. I felt that too
much time was spent in recapitulating old topics and bringing
new workshop participants to the point reached at the end of
the last meeting. On some of the old topics (e.g., interactions)
we did not even reach that point. For this reason, we did
not have enough time to get far into new topics. Kenny Crump's
working paper on less-than-lifetime exposure was very helpful.
By contrast, we needed stronger working papers on other topics,
such as interspecies conversion of dose and reproductive toxi-
cology.
I think it is over-optimistic to expect to reach "consensus"
on topics at or beyond the frontiers of research, such as multi-
chemical assessment. I understand that ECAO has to adopt a
set of procedures and needs support from the leaders of the
scientific community. Nevertheless, several scientists at
the workshop expressed "scientific" caution in making generaliza-
tions. I would have liked to see some informal voting on some
topics, with the scientists being required to vote for one
procedure and not being allowed to avoid expressing a preference
by abstaining from voting.
A-52
-------
Memorandum
Page 2
August 25, 1983
Interspecies Conversion of Dose and Duration of Exposure
The discussion of interspecies conversion of dose was
chaotic because several participants failed to define their
units. Thus, some participants referred to a dependence on
two-thirds power of weight, while others referred to a dependence
on one-third power of weight. Both mean the same thing, because
the first is appropriate when applied to dose in mg, and the
second is appropriate when applied to dose in mg/kg. Before
consensus is reached, I think that ECAO will have to commission
a well-thought out paper by someone who really understands
the mathematics of allometric relationships (few, if any, of
the speakers appeared to do so). In the meantime, I would
vote for a general assumption that the effective dose in mg/kg
scales as the one-third power of body weight.
As I tried to point out, scaling in proportion to ppm
in the diet is only appropriate if diets of all species are
comparable (all dry weight or all wet weight, not a mixture
of both). No one seems to know the dry weight of a human diet.
The discussion of conversion of dose for duration of expo-
sure seemed confused. Scaling duration in proportion to species
lifetime seems appropriate for carcinogenesis, but not for
other toxic effects. I am not convinced that a 90-day exposure
of a rodent is equivalent to a 9-year exposure in humans.
For acute, subacute, and subchronic exposures, I would vote
for scaling in proportion to biological half-life, but I am
not fqre this is independent to the scaling in proportion to
W~0,33, since both reflect differences in metabolic rate.
I think that more empirical information is needed, both for
humans and animals.
Health Risk Assessment for Less Than Lifetime Exposure
I liked the paper and presentation by Crump, which makes
a lot more sense than scaling in proportion to cumulative expo-
sure. I have already been using essentially the same formulas
(based on papers by Whittenmore and Day & Brown) for carcinogenic
risk assessment.
For noncarcinogenic effects, I think the empirical approach
exemplified by the chart for methoxychlor is appropriate, but
I think this needs more work. Kenny Crump and I undertook
to work on statistical approaches to analyzing such data.
A-53
-------
Memorandum
Page 3
August 25, 1983
APIs Based on Quantal, Continuous, or Graded Data
QCG data can always be converted to dichotomous data by
setting a more or less arbitrary criterion for the level at
which a change becomes an "adverse effect." However, this
throws away a lot of information available in the original
data.
For truly QCG data, I think the concept of ADI is inappro-
priate, since such data can often be fitted to nonthreshold
models. As for dichotomous data, most toxicologists are concep-
tually confused between intra-individual and inter-individual
contributions to the dose-response curve. When inter-individual
variability is substantial, the "equivalent background" or
"dose-additivity" concepts invalidate the assumption of thresh-
olds. We need better thought-out models.
Route-to-Route Conversion
1 thought we approached consensus that systemic dose was
an appropriate method of conversion for systemic effects, but
that no method of conversion was appropriate for effects at
the site of first contact.
Multiple Chemical Assessment
I don't think we got any further with this difficult topic
than we did at the last meeting. At that meeting, I was almost
convinced by Crump's argument that to first order, doses of
different chemicals should be additive. Dose additivity is
the key concept and needs empirical investigation.
Structure-Activity Relationships
I was impressed by Enstein's presentation and regret that
there was not enough time for him to explain just what was
being analyzed in his multivariate analyses.
Use of Reproductive Effects as Endpoints
I thought Jeanne Manson's presentation was too much con-
cerned with detail and did not approach the topic of whether
reproductive toxicity data can be used for risk assessment.
I strongly disagree with her implied negative on this question.
I recently published a book on this topic (I.C.T. Nisbet and
N.J. Karch, Chemical Hazards to Human Reproduction, Noyes Data
A-54
-------
Memorandum
Page 4
August 25, 1983
Corp., 1983) in which we explored the concordance between human
and animal data for the same chemical and cautiously concluded
that a semi-quantitative concordance exists if one can identify
the appropriate animal species. A copy of the summary of this
study is enclosed.
Use and Biological Justification of Mathematical Models
I don't think we progressed significantly on this topic.
High Risk or Sensitive Subgroups
1 think that there is abundant evidence for a wide range
of susceptibility within the human population to a number of
toxic chemicals. However, I don't think we have many cases
where we can assume that the distribution of susceptibilities
is bimodal. (If we did, we could no longer assume that a linear
dose-response curve is conservative, since a bimodal distribution
of susceptibilities implies an S-shaped dose-response curve.)
For most effects of concern in the human population (including
cancer, reproductive, and neurobehavioral impairment) there
is a substantial background frequency, suggesting (but not
requiring) the concepts of equivalent background, dose bdditivity,
and linear nonthreshold dose-response relationship. These
ideas need thoughtful study for effects other than carcinogenesis.
I made the point that small children are more active,
spend more time outdoors, eat more, drink more, and breathe
more (per unit body weight), and have some presumptively more
sensitive organ systems than adults. Hence, the conventional
safety factor of 100 is never sufficient to protect small children,
since it includes a factor of only 10 for intraspecific variability
in sensitivity.
Use of Exposure Data in Health Risk Assessment
Despite the presentations at the last meeting, most partici-
pants at the workshop seemed to think that exposure assessment
was a straightforward problem in measurement and modelling,
and could be carried out independently of hazard assessment.
In the few minutes available for discussion, I tried to point
out that exposure assessment is extremely difficult with the
limited resources usually available at hazardous waste sites,
and usually is the critical factor limiting the accuracy of
health risk assessments.
ICTN/mrs
Enclosure
A-55
-------
El 1 en J. 0'Flaherty
Postmeeting Memorandum for
"Risk Assessment of Multichemical Exposures" Meeting,
EPA, Cincinnati, July 12-13, 1983
Consensus Topics
A. Interspecies Conversion of Dose and Duration of Exposure
Consensus: For interspecies conversion of dose, continue with EPA's pre-
sent procedure in which the ratio of equivalent dose rates in two species is
considered to be inversely proportional to the cube root of the ratio of the
mean body weights of the two species. However, there seemed to be some feeling
that this is a default position; that is, 1t is reasonable and has some experi-
mental support, but nonetheless our understanding of species equivalence of dose
rate is incomplete at present. For duration of exposure, the consensus was to
continue to use fraction of lifespan for compounds that produce chronic effects,
but that time to development of acute effects may be relatively constant across
species.
With regard to interspecies conversion of dose or dose rate, I believe that
sufficient information is available in the published literature to enable us to
address this question much more systematically than has been done by the EPA to
date. Harold Boxenbaum has recently (1982) published an authoritative review of
interspecies scaling of pharmacokinetic behavior. This paper would be an
excellent starting point for an analysis of information on effects across spe-
cies. Two questions that need to be addressed specifically are: Should
interspecies scaling be carried out differently for short-term
(non-steady-state) and for chronic exposure situations? How should interspecies
scaling be carried out for compounds that are activated, rather than detoxified,
by metabolism?
B. Health Risk Assessment For Less Than Lifetime Exposure
Consensus: For noncarclnogens, continue to develop the dose (mg/kg/d)
-duration (fraction of lifespan) plot, working toward Incorporation of greater
objectivity into the process of determination of NOAEL and ADI lines. For car-
cinogens, use of the multistage model as proposed by Crump and Howe was endorsed
except perhaps for childhood leukemlas.
I agree with both positions and have no further comment on this point.
C. ADI's Based on Quantal, Continuous or Graded Data
Consensus Fit the entire data set, whenever a data set exists, to obtain
an estimate of something or other.
This consensus was not at all clear. I do not believe that most of the
participants had thought through the implications of carrying out this procedure
with both continuous and quantal data sets. In fact, my own position on this
issue has moderated somewhat since the meetings as I have given it further
thought.
A-56
-------
The general approach, which is to apply various uncertainty factors to a
dose selected on some defined basis, will not be altered. The uncertainty fac-
tors used by the EPA not only have a history of successful (i.e., successfully
protective) use, but they have also been given a measure of at least empirical
respectabi1ity. In particular, application of a factor of 10 to account for
within-population variability appears to be adequate for most chemicals.
The current procedure for selecting the dose to which the uncertainty fac-
tors are to be applied is something of a hybrid. Graded toxicity data are used
to establish the presence (often at an unspecified response level) or absence
(zero observed response) of an effect 1n a test group of animals. I have always
interpreted the various "effect levels" used by EPA as doses associated with
some, perhaps unknown, quantal response or occurrence of the effect.
Interpretation of one of the uncertainty factors of 10 as accounting for within-
population variability is, thus, logical.
Fitting a quantal response data set with a model so as to utilize all
available Information thus causes no conceptual difficulty. My preference would
bes to calculate a confidence limit on the dose corresponding to a specified low
risk, say 10-2, and then to apply the appropriate uncertainty factors. This
approach is relatively model-independent.
If the recommended procedure, of fitting all the data points, is applied to
a continuous data set, then presumably a confidence limit will be calculated on
the dose corresponding to an acceptable meann effect intensity. Translated into
the language of quantal response, this means that 1f the distribution of magni-
tude of effect intensity among members of the population is approximately nor-
mal, the dose corresponding to an acceptable mean effect intensity is also the
dose associated with appearance of a magnitude of effect greater than or equal
to the mean in about half the population; that is, it is an ED50 with respect to
the acceptable effect intensity. If the distribution of magnitude of effect
intensity among members of the population is significantly different from a nor-
mal distribution, then the dose associated with the acceptable mean effect
intensity Is not an ED50.
To sum up, four distinguishable methods have been used or are proposed, to
select the dose to which uncertainty factors are to be applied. These are:
Type of Data Starting Point for Application of Uncertainty Factors
1. NOAEL or NOEL Dose associated with zero quantal response in experimen-
tal group.
2. LOAEL Dose associated with a positive quantal response, frac-
tion affected often unknown but may approach 1.0, in
experimental group.
3. Fit to quantal Dose associated with a low quantal response of arbitrary
(response) data magnitude, say 10-2.
4. Fit to continuous Dose associated with an acceptable mean effect inten-
(effect) data sity. Quantal response unknown unless distribution of
effect intensities is known, but probably is of the
order of 0.5.
A-57
-------
If one of the uncertainty factors used by the EPA is to be considered pro-
tective for population variability, it must logically be applied to dose-quantal
response data. All the current or proposed methods use as starting points doses
that can be associated with quantal responses as shown above. However, these
quantal responses vary from 0 to 1.0. Selection of uncertainty factors to use
with these methods should take this variability into account.
D. Route-to-Route Conversion
Consensus: Route-to-route conversion may be undertaken, if necessary, for
compounds with long half-lives and insignificant first-pass effects. However,
it was generally agreed that lung-to-gut extrapolation based on a TLV 1s pro-
bably not advisable, since TLV's are likely to have been based on the irritative
properties of the chemical rather than on its systemic toxicity.
In general, I am in agreement with this position. However, certain refine-
ments were discussed during the workshop session on multiple route exposures. I
include a summary of the conclusions and proposal of that workshop under the
workshop heading.
E. Multiple Chemical Assessments
Consensus on this topic was not reached, in my opinion. Most of those pre-
sent seemed to believe that estimation of a total hazard index, using the prin-
ciple of dose additivity, 1s the least unreasonable method to use in the absence
of specific information concerning the chemicals' toxicities and interactions.
However, some participants, particularly those who were present for the first
time, clearly were not comfortable with the idea of estimating total hazard at
all.
I agree that estimation of a total hazard index based on dose additivity is
the most defensible method at present for evaluating health risk from exposure
to a mixture of chemicals. I support this approach for two reasons: (1) it is
conservative in that, of the various possible methodologies, it is the method
resulting in the largest estimated hazard associated with a mixture of chemicals
at known concentrations: and (2) it takes into account a contribution by all
toxic chemicals known to be present at the site.
Reliance on a "marker" (most toxic) chemical 1s of no use in estimation of
total hazard unless (and until) it can be shown that the kinds and relative pro-
portions of chemicals present at waste dump sites are reasonably constant across
sites. Even then, reliance on a marker chemical for hazard estimation would be
risky, since the presence of a large amount of an only moderately toxic chemical
could greatly increase the actual total hazard. If a marker were used for
hazard estimation, other toxic chemicals probably would not actively be sought
at the site. Use of a marker chemical to screen sites for classification and
for further study is subject to the same uncertainty; however, use of a marker
chemical for initial screening purposes may be justifiable on the basis of speed
and ease. If this is done, the possibility that a site judged to be a low-risk
site may not in fact be low-risk, must be borne in mind.
Workshop Topics
I attended only the workshop on multiple route exposure. Discussion at
this workshop, although it was moving in the direction of consensus, did not
A-58
-------
lead to a definitive statement by the end of the session. Consequently, Dr.
Clarkson, Dr. Pepelko and I, with input from Dr. Hartung, spent the time sche-
duled for the second workshop session in formulating a summary statement based
on the discussion at the multiple-route workshop. My recol 1 ectiorj of the key
points follows.
Summation across exposure routes is possible in principle. The simplest
summation method is
= ^lrl + ^2r2 + •••• + c'ir1 »
where dn = total dose, dj, d? d-j = the doses presented to routes 1, 2,...,
i; and rj, rj...., r-j = the fraction absorbed for each route.
The Ideal chemical candidate for exposure summation across routes would be
one that meets all five of the following criteria, outlined by Dr. Hartung:
1. Internal dose-duration of exposure patterns are identical for all
routes of exposure. (Thus, for example, concentration "spikes" in the blood do
not occur for one exposure route while smooth buildup to a steady state blood
concentration is observed for a second exposure route.)
2. Systemic effects are identical for all routes of exposure. (Systemic
effects might not be identical for all routes of exposure if, for example, there
were a first-pass effect for one route that produced a metabolite whose toxic
effects were different from those of the parent compound.)
3. There are no route-specific effects at the portal of entry.
4. There are no first-pass effects.
5. The relationship between applied dose and dose delivered to the target
organ is known. (Internal dose, or dose related to concentrations in the blood,
is usually considered to be equivalent to delivered dose. However, it may not
be.)
Since the characteristics of few compounds will be known with this degree
of completeness, it may be possible to state some generalizations about the
extent to which different groups of compounds may meet these criteria:
1. Internal dose-duration patterns are likely not to differ significantly
across routes of exposure if the chemical has a long tfy. Concentration cycling
around the mean steady-state blood concentration is damped 1f the compound has a
long tfo, so that for such compounds the difference between dose-duration pat-
terns Tor intermittent-exposure routes and for continuous-exposure routes is
minimized. Alternatively, even for compounds with short t]M, if patterns of
exposure are roughly comparable the internal dose-duration patterns may also be
roughly comparable. For example, exposure through water, although discon-
tinuous, would fall somewhere between continuous exposure by inhalation and
administration of the same total dally dose to an animal by single gavage.
2. Although situations in which systemic effects are different for dif-
ferent routes of administration can readily be imagined, we were unable to iden-
tify any compounds for which this is true. It ma^ be possible to infer that few
chemicals do not meet this criterion.
A-59
-------
3. Highly chemically reactive compounds, or compounds with long residence
times in the tissue of entry, should be suspect with regard to their potential
for causing portal-of-entry effects. However, it cannot be assumed that com-
pounds that are not highly reactive or have short residence times in the tissue
of entry will not cause portal-of-entry effects.
4. Compounds with long tare less likely to undergo significant first-
pass effects than are compounds with short t]^. Thus, as a result of both con-
siderations 1 and 4, it appears that compounds with long ty^ may be suitable
candidates for summation across exposure routes.
The assurance with which summation across exposure routes can be carried
out varies both with the nature of the data set(s) on which the summation is
based and with what is known about the degree to which the compound meets cri-
teria 1-5.
A-60
-------
POSTMEETING MEMORANDUM OF MARVIN S CHNEIDERMANN
This report relating to the EPA-Cincinnati meeting of July 12-13, 1983
consisted of these parts:
1. Continents on the "consensus" portions of the meeting (1st day).
(Greatest attention given to IA-. Where Issues overlap, they are
largely discussed under IA.)
2. Summary of workshop A1 (Sensitive sub-groups)
3. Comments on other workshops
4. Suggestions for future meetings (if any). The signal R-» shows
where a specific new recommendation is made.
IA: Interspecies conversion of dose and duration of exposure.
A. There are two major areas that remain to be resolved:
1. The biological or mechanistic basis for the use of any
transformation. The surface-area transformation seems to be
somewhat more useful than the mg/kg transformation - despite
the finding by Meselson several years ago that for the six
chemical carcinogens for which there were satisfactory animal
and human dose data, the mg/kg transformation provided the most
satisfactory interspecies comparison.
If the specific toxicity associated with a material is related
to the metabolism of that material (as, for example, in "first-
pass" effects) then it appears reasonable that the interspecies
conversion factors are likely to be effect-specific. That
might imply that Meselson is correct for cancer as a toxic re-
sponse, while the 2/3 power or surface-area transformation is
appropriate for some (many?) other toxic effects. This, in
turn raises regulatory Issues relating to the severity of
effects, implying that perhaps the appropriate transformation
is the one that yields the lowest "acceptable" dose for the key
or crucial (or "swing") toxicity.
2. The different metabolic rate in children may require a differ-
ent species-conversion or transformation than for adults. If
it is true that children have greater intake (food, air, water)
than adults per unit weight and/or per unit surface area, then
(if different conversion factors are not appropriate) it is
likely that tighter standards may have to be set for childhood
as opposed to adulthood exposures. This issue is part of the
concerns of workshop Al- high risk sub-groups.
A-61
-------
B. Other concerns: combination of "toxicities."
The consideration of multiple exposures (as in toxic waste dump
effluents), and the combination of exposures via different media is
still unresolved. In particular, it seems to me that a set of
intermediate guidelines needs to be set out now (to be revised, as
necessary, in the future) on:
a. Multi-media exposures, including when and how combined
exposures can (should?) be taken into account, effects of
"dominant" exposures, appropriate pharmaco-klnetic effects when
known (and how these will get to be known better in the future).
b. Combinations of sub-ADI exposures. (A sophisticated way of
handling these may be able to subsume the multi-media exposure
problem - If AOI's are, or can be, set up for each route of
exposure).
R At the moment I am inclined to recommend summation of
EXPOSURE/ADI for similar (same?) toxic responses, and then
either a grand summation across all toxic reponses (which data
I doubt exists) or an evaluation of the first stage summation
for the key, crucial, dominant, swing toxicity (much of which
data will also be missing).
R -» A system of "default" penalties for missing data will probably
have to be developed. The absence of toxicity data should lead
to more rigid standards for exposures In general. (I can
conceive of circumstances in which data on a serious toxic
effect would make it unnecessary to develop data on a less
serious effect, however). Not knowing should carry more of a
penalty than knowing. A possible "penalty" system would assign
an upper 95% toxicity level (as found in the distribution of
all materials that have been tested for the specific toxicity)
to any untested material.
e.g.,:
Distribution of ADI's for Toxicity "A"
for all materials tested to date
ADI (dose)
ADI arbitrarily assigned to any untested material
C. Summary
1. For use until future notice:
A-62
-------
a. Surface area transformation
b. ppm (in food, water, etc.): no transformation
(Be concerned about adult-child differences)
c. Same transformation for all toxicities
(Be concerned about adult-child differences)
d. Same transformation for all toxicities
(Be concerned about pharmaco-kinetics)
e. Same transformation (possibly into ppm) for all media
f. Penalize missing data
(develop a system to be presented at a future meeting)
g. Summarize total exposures by
C H 5Z I Exposure)
ADI /
ijk
i ¦ materials
see
j ¦ toxicities l....n£ text
above
k ° media 1....n^
2. Research Issues (a sampling)
a. (as indicated above)
b. Effect of one time exposure
- spills, in contrast to elutlon over time from dumps
c. Effect of changing exposures with time
- mechanism of toxicity (i.e., relation to cumulative
dose? peak dose? etc.)
d. Differences among materials that need to be
activated
materials that are direct-acting
materials that are inactivated.
Session IB.
Health risk assessment for less than lifetime exposure
A. Mathematical modelling
1. Recognize that few disease-dose models exist for other than
cancer. I am not certain that models really need to be
developed for all (or even for many) of the non-cancer
effects. Models are needed only there Is uncertainty about
A-63
-------
thresholds and/or the interactions with other exposures
producing the same effects (hence the concern about dose
additivity vs response additivity). At present I see a
possible need for models for chronic disease effects other than
(or in addition to) cancer. (For example, for arthritis,
emphysema, hypertension, other coronary/circulatory diseases.)
2. Mathematical modelling for cancer is developing greater
sophistication. The major missing link that I see is in
bringing two major classes of models - the "sensitivity" models
(logit, probit, Mantel-Bryan, tfelbull, time-to-tumor) and the
"mechanism" models (multistage, multi-hit, one hit) together,
giving us a mechanism model that also takes sensitivity into
account. At present, I find the multi-stage model as the most
reasonable, reflecting both the laboratory knowledge
(Berenblum-Shubik, initiator-promotor) and the epidemiology
(Armitage-Doll). It seems to be deficient in explaining at
least two, phenomena, however:
a. Childhood cancers, particularly childhood leukemias
b. Declines after an initial increase in (absolute)
Incidences of some forms of cancer (following some
industrial exposures), including some lymphomas.
An exhaustion of a "sensitive'' population may have to be
invoked for the latter phenomenon. I have some concern about
"virulence" of cancer as possibly related to dose or exposure.
I know of no data on this.
2. NOEL-ADI (Dourson - presentation).
a. This is an interesting idea, worth follow up and further
development. Two things, in general, are missing:
i. Use of dose-response relationships
11. Use of information about sample size (Charles Brown
of NCI has shown that for small sample sizes the
highest dose (d3) in one experiment in which the
true responses were: dj^".01, d2 ™ .05, d3 ¦
.10, also had the highest probability of being
selected as the NOEL. It was not until the sample
size was in the vicinity of n"80 for each dose and
controls, that the probability that the NOEL lay
below dj, was greater than 0.5.)
As long as NOEL's are defined in terms of "statistically or
biologically significant" increases issues of sample size will
be important. (Does an objective, agreed-upon definition of
"biologically significant" exist?)
b. The "bench-mark" approach suffers some of the defects one
has in cancer-modelling, i.e.
A-64
-------
i. Choice of bench mark level, e.g., .1, .01, .001
11. Dose-response model for extrapolation downward - what
model? If a model with a threshold, how Is threshold
arrived at? If a threshold, hasn't one then
arbitrarily defined the NOEL? If no threshold, then,
no NEL, even If one could argue that there was an
experimental NOEL.
3. Cancer-modelling - early stage, late stage effects (Crump)
a. This is a very useful extention of the Day-Brown,
Whittemore-Keller work based on Armltage-Doll. I have a
few general comments, and a few specific comments.
i. The animal studies conducted to date seem Incapable
of describing more than 2 stages, initiation and
promotion. (Although I have seen one study which
describes two stages within "promotion.") I would
recommend some joint work between EPA and NTP to
design animal models capable of distinguishing more
stages - and at what stage(s) the carcinogen is
operating. It must be recognized, in addition, that
the animals used are usually genetically more
homogeneous than humans and may, as a consequence
show fewer (apparent) stages than would humans.
11. I would like to see Tables 1 and 2 (and related
tables) redone to show the effects of a "remaining
lifetime" exposure resulting from modifying a dose -
rather than showing a fixed numbers of years effect.
For example I would like to see a table giving
effects of exposures d from age 0 to ages 20, 40, 60,
then followed for the remaining lifetime (use U.S.
life tables - say it would be 55 yrs, 37 yrs, 21 yrs
- or something like that) to a dose d/10, d/100.
This would mimic the situation that EPA is faced with
in reducing exposures from d to some d/q - which, it
is assumed will then not be exceeded for a remainder
of a life-time. Some other, probably more realistic
life-time exposure patterns could be decided upon-
and the table constructed for these. 0SHA (or NI0SH)
should have some Interest in this, too, as shows
CPSC. (Question: Does N. Breslow's paper In Jour.
Am. Stat. Assn., March 1983, have any bearing on this
problem?)
ill. I would like to see some parallel tables for
materials which appear to act at more than one stage
- e.g., cigarette smoke, and, if we keep all forms of
cancer together, asbestos. Some animal data might be
useful here, too- e.g., 2-AAF in the ".01 study" -
with both bladder cancer and liver cancers.
A-65
-------
iv. Heuristically, an association between early stage,
late stage and initiator-promoter, and "shown
mutagenic", "not shown mutagenic" can perhaps be
made. I would suggest collaboration work between Dr.
Crump and some EPA/NTP/NCI biologists to work out the
appropriate ways to say this.
v. Some details:
(1) The two parts of the Tables 1 and 2, labelled
"same total doses" and "same dose rate" are only
expressions of different cumulative dose. In Table
1, the "same dose rate" piece reflects half the total
dose of the "same total dose" piece (i.e., 35/70).
In Table 2, the relative dose is 2/7 (i.e., 20/70 -
.2857).
(2) I do not understand (in conceptual terms) the
apparently opposite direction of certain effects -
e.g., Table 1:
First stage only
Age beginning
0
20
STAGE
4 5
177517881794
0.71 0.52
0.37
1.97
0.26
Note
Upward trend
by stage
Downward
trend by stage
Penultimate
stage only
20
35
1.0
1.37
1.0
0.62
1.44
1.38
0.38
1.36
1.62
0.22
1.22
1.78
Downward by
stage
Up and down
Up by stage
Is there some way to make (or help) a person understand how/why
these different directions in trend are reasonable?
vi. The determination of the number of stages and which
stages are affected seems to very critical. Is there
some objective way to do this? (See my earlier
suggestion re: animal experiments) there seems to be
a related problem in the animal experiment (EDB)
analyzed by Crump. There a distinction needed to be
made in tumors causing death and tumors discovered at
A-66
-------
time of death as an incidental, non-fatal event, (see
Crump p.13)
Finally if one postulates some number of stages and
argues that the carcinogen operates at only one of
the stages, is there any (intellectual) necessity for
explaining what produces effects at the other stages-
i.e., what makes the other stages come about?
vii. Comment on Table 4:
The hundred-fold differences between the week 49
stage 1 dose (4.7 mg/kg?) and the week 105 stage dose
(0.048) is of considerable consequence when looking
at possible human experience - where about half of
all cancer deaths occur in persons older than 65.
viii. Page 19 of Crump.
The speculation on the reasons for non-linearity is
very important, implying that pharmacokinetics needs
to be investigated rather thoroughly. A lot of work
needs to be done on this possible reconciliation.
Because of the peculiarities in trend (Tables 1 and
2) with stage, e.g. if stage is not the source of
non-linearity, then very different "safe" doses are
likely to be recommended.
SESSION 1-C
Some comments Included above under "bench mark" considerations. My
general impression is that using graded, or continuous responses (which,
in effect, uses more of the data than does using quantal responses) will
lower NOEL's.
SESSION I-D
Some comments Included above under Session IA. No further comments -
other than indicating relationship to Crump speculation on non-linearity.
SESSION I-E
Little new material was presented. See my comments on additlvity of
EXPOSURE/ADI ratios for similar effect. For cancers, either confine
this summation to cancers at the same site - or, with somewhat more
sophistication- to cancer arising out of tissues of similar embryonic
origin - perhaps all carcinomas, all blood-forming tissues (plus
lymphomas?), all sarcomas. Advice from pathologists should be sought.
John Berg (Iowa) has done some work along these lines while considering
multiple primaries.
A-67
-------
SESSION II-A Structure-activity relationships
This is an area of promise, and needs to be supported. The only other
claimed to a useful SAR approach that I know of is the electrophile
approach by the Millers of Wisconsin. Enslein's results should be
compared with the Millers' - and the reasons for disagreement (I am sure
there are disagreements) worked out- if possible. I would like to see a
detailed exposition of the Enslein process including:
1. The discriminant equations
i.e., what parameters contribute most to separating materials
later labelled as carcinogens from those labelled
non-carcinogens? Are these parameters different for different
classes of chemicals? If so, how are classes of chemicals
defined, or determined in advance?
2. What kind of self-learning process Is built into the Enslein
procedure? How have the parameters changed with additional
data? How are missing data handled?
3. Comparisions of Enslein equations with other SAR approaches-
and with such testing procedures as short-term mutagenesis
tests. This would (or should) include a proposal for step-wise
testing (evaluation) of potential carcinogens. Can the Enslein
process be set up with cut-off points that we can be sure will
not be carcinogens? If this could be done, a great deal of
testing might be eliminated. Some cooperation with NTP seems
In order here.
Numerical comment. I believe there is a theorum in
discriminant analyses that says the discriminating power of
any discriminant equations must be lower for every set of data
in comparison to the data set from which the equation was con-
structed. (A corollary to this is that the discriminant
equation will change, I.e., will show the effect of "learning,"
as more materials are examined.)
Concerning the question of why a black box (a discriminant
equation) works, it is worth noting that the black box Is not
created from random elements, but rather that the parameters of
the discriminant equations are chosen because someone thinks
they have something to do with the outcome. Frequently much of
the explanatory information is extraneous, and contribute
nothing (additional) to the correct sortlng-out process.
Final question: Is there some way of adding the electophile
Information to the existing Enslein equations? If done, does
it improve the discrimination?
A-68
-------
SESSION II-B Reproductive effects
This area needs to be broadened to evaluate other effects If It is to be
useful for assisting in the regulatory process. Specifically there
needs to be consideration of pregnancy wastage, (spontaneous abortions),
reduced fertility, etc.- as well as teratologic effect, malformations,
etc. This impresses me as a nascent field, still in the descriptive or
quantal phase. It is nonetheless useful.
To me these developments are possible:
1. Establishment of a minimum body of end-points to be looked for
in these studies
2. Creation of data bases on spontaneous abortions (in humans) -
and subsequent interaction between the epidemiologic and
laboratory studies- including cooperation on "behavioral"
toxicology.
3. Studies on effects on the germ cells in whole, intact, mammals.
SESSION II-C Mathematical models
Much of the discussion under IB, IC is related. Response models as well
as dose models need to be considered. I see needs for considering:
1. Effects of peak doses versus effect of cumulative doses (there
are pharmacokinetics issues here)
2. Population variability- e.g., the population distributions of
t, in the equation (by Hertzberg)
Dx < (t1A(t1)-D0to)/(t1-to) to £ ti $ 1
where l'llfetime
3. Tissues-specificity problems - both arising out of first-pass
considerations and other mechanisms
WORKSHOPS:
A-l Considering high risk (sensitive) subgroups in health risk
assessment.
(What follows is the report of this workshops.) The group came
to these (agreed-upon) conclusions.
1. One "safety-factor" number (i.e., division by 10 or 100) is
Inadequate to take into account the variability in the human
population. Safety factors developed for adults are
inappropriate for use with children.
A-69
-------
Action: Develop more data relating to safety-factors for
different segments of the population. In the Interim to use
current factors, setting a deadline for when they should be
reconsidered.
2. For future changes In safety factors, base safety factor on
(more) data on:
a) Slope of the dose-reponse curves (The shallower the slope
the larger the factor needs to be)
b) The specific toxicity and its severity (the greater the
need to protect against the toxicity, the large the safety
factor needs to be. A life-time of illness or reduced
functional ability is usually more important than a
short-term, reversible, Incident).
3. For non-cancer effects, remain with the ADI concept, rather
than a "hazard index". While the public-health consequences
are greater when more people become ill (hazard index), it is
individuals who become ill, and even a small number of persons
are to be protected against serious consequences.
By way of back up of these agreed-upon conclusions the participants
considered these questions (stimulated by Erdrelch presentation):
1. What contributes to sensitivity in addition to demographic
factors?
•
a. Prior exposures to same, and other materials
Genetic background
Existing other illnesses
Special dietary sub-groups
Data are largely confined to drugs and rarely exist for non-lntented
effects, although some interaction data with common human activities
(i.e., exposure to alcohol, cigarette smoking) do exist.
Some data exist on humans- measured, controlled, consistent among
Individuals.
2. What resources exist to desrlbe the variability?
(Data bases or follow-up groups)
A. Registries of special diseases or conditions, e.g.,
bronchitlcs, persons with transplants, cancer registries:
Framingham
Michigan
Fels Research Institute data- for follov-up of some children's
groups
A-70
-------
Special exposure groups
-industrial
-Love Caval
HANES
NIH-NICHHD- peri-natal follow-up group
Retirement (e.g., Leisure World) populations etc.
3. Should chemical-specific lists of high-risk groups be established?
A. Yes, and the lists should be cross-indexed by effect, and
sensitive group as well as by material,
e.g., Phenylketonurlcs (aspartame)
Asthmatics
Persons with angina
Alcoholics etc.
4. Is pharmacokinetic information (existing data) partially useful in
this area?
A. Yes, but recognize shortcomings- not a direct measure of health
effect, e.g., more rapid or Intense response in "sensitive"
persons. Factors related to receptors- likely to be genetic,
but nonetheless chemical-specific (One must not assume
uniformity in a population with respect to behavior of
receptors)
The participants also reviewed some issues varied by Hattls. He
outlined these contributions to variability:
Non-chemical specific: cross-sectional studies of humans likely to
provide some useful data.
a. Things causing variability In exposures (often route specific):
Body size - see children - greater water Intake per unit
weight or surface
b. Overt disease states- pushing persons closer to a thresholds
Angina
Asthma
Nerve-conduction defects (increasing reaction time)
c. Physical inhibition (or excitation) of function by prior exposure:
Alcohol (liver function)
Smoking (hemoglobin availability).
More likely chemical specific- requiring studies of individuals
Pharmacokine tlc
Parmacodynamic
[End of this workshop report]
A-71
-------
Workshop A-2
I see three serious Issues:
1. The rating values for toxic effects have fuzzy boundaries
- reversibility seems not considered
- germ plasm effects?
- life-time effects (e.g., reduction in functional ability,
reduced IQ)
2. Dose producing effect
- perhaps, rather than in absolute units, may be reported as a
proportion of the anticipated dose to humans
3. Scoring system implies a logarithmic scale (equally spaced in
the log) for both effect and dose. This needs a lot of
justification. The boundries for RQ's are arbltary. Some
justification must be made.
This process Is far from ready to be used for extrapolation to NOELs.
At this stage of development, I can see a lot of controversy, and only a
little gain.
SUMMARY SUGGESTIONS FOR FUTURE MEETINGS
I found the two workshop sessions I attended as the most useful
expenditures of time. This leads me to suggest:
1. Keep future meetings small
2. Have workshop sessions on first day - from which a written
report is made to the whole group - and is made available
by the next day
3. Have a pre-set format for the report, and a set of questions
for the workshop to answer. Assign a specific (contractor?)
person to assist the rapporteur in preparing each workshop
report, so that the reports might be complete. I think each
report might be required to address the following:
A. Action recommendation - what to do now, and
up to such time as further data become available
B. Appropriate additional research to answer
C. Unresolved issues of substantial consequence.
4. Have the whole (small) group meet on the second day to consider
the workshop reports, to make necessary changes, etc. - leading
to
A-72
-------
5. A final report to EPA with recommendations (in the same format
as the workshop reports) to which the whole group will agree-
or to which some members will prepare minority reports.
A-73
-------
ENVIRONMENTAL DEFENSE FUND
MEMORANDUM
TO: C. Cueto, PhD ' (?] \
FROM: E.K. SILBERGELD. /''v ; r*i J
RE: Risk Assessment Meeting *
DATE: 18 July 1983
This constitutes my postmeeting memorandum, and follows the
outline as indicated in your memorandum of 11 July.
As a new participant in this ongoing discussion of risk
assessment, I would have found it very useful to have received
more premeeting briefing material on the past sessions of this
group and on the progress of its earlier consensus. It took
about half of the first day for me to understand the structure
and process of this group. When I did, I came to appreciate
the method highly, but it is a difficult one to jump into
without prioc knowledge.
As a second general comment on this workshop, I would like
to express my overall concern (which will be detailed below)
that we failed to deal with noncarcinogenic toxicants, despite
the topic headings. There are significant differences in
carcinogenesis, a6 compared to other types of toxicology, in
the types of data available as well as the appropriate models
which can be used to extrapolate from datapoints in order to
estimate effects at lower doeses. This discussion group needs
to focus more specifically on noncarcinogenesis toxicology.
CONSENSUS TOPICS
I. Interspecies conversion of dose
and duration of exposure
ECAO wants an approach which provides the most appropriate and
inclusive use of data derived from a range of experimental
animals in order to estimate risk to humans. There are
A- 74
1525 18rh Street, NW Washington, DC 20036 202«833*1484
OFFICES IN NEW YORK NY (National Headquarters), WASHINGTON, DC, BERKELEY. CA. DENVER. CO
-------
problems related to differences in body size and in relative
lifetimes. I was not present for this discussion during the
general presentation on day one; however, I gather that
consensus was generally reached on the use of a dose/body
surface area or cube root of weight approach, since those
method give some accomodation to organ:body weight ratios and
have some relationship to metabolic rate (e.g., liver:body
surface area).
My concern with this approach is that it is relatively easy
to think of examples which confound any generic approach--e.g..
the qualitative differences in caffeine metabolism between
humans and rodents. However, there is obviously a need to
develop generic approaches, particularly when data are not
available. My own research in pediatric pharmacology has been
based on surface area adjustments, so this approach has some
clinical justification.
II. Risk assessment for less than
lifetime exposure (toxicants)
The consensus appears to have adopted the usual pharmacological
conventions of 0 -5 days as acute exposure; 5 -14 days as
subchronic; and 90 days as chronic.
I did not hear a discussion of comparing physiologically
similar life stages: e.g., the first 6 days of rat gestation
with the first trimester of human pregnancy, or the rat's first
20 days of postnatal life with the nursing period of infants.
III. ADIs based on quantal. continuous
or graded data
This session proposed a rather fundamental revision in the way
in which health risk assessments are to be done. Although the
title of the session was directed towards noncarcinogenic
toxicants, most of the presentation by Crump, and the
subsequent discussion, was focussed on carcinogens. The
proposal by Crump was that instead of applying models (such as
Mantell-Bryan; Weibull, Armitage-Doll. etc.) in order to draw
dose-response relationships beyond the range of the data, EPA
should only U6e dose-response data within the range of actual
experimentation (or epidemiological observation). Also,
instead of working to establish NOAEL points through
experimentation, research should concentrate upon the
definition of "benchmark" (or, in Nelson's term. pseudo-NOAEL.
[2]
A-75
-------
although I think this is a misinterpretation of Crump's
proposal), and then apply appropriate safety factors or other
divisors in order to define an acceptable level.
I do not think that consensus was reached on this topic.
For one thing. Crump's paper was not easy and was not discussed
in depth at the meeting. For another discussion (such as it
was) focussed only on these hypotheses as applied to
carcinogens. Crump did not have time to discuss his models for
noncarcinogenic toxicants.
More importantly, I have grave reservations about this
approach. I fear that it really begs the question. While I
endorse the concept of working from defined effects, rather
than nebulous and often highly debatable NOAEL doses, I think
that the way in which one would "work down" from such
benchmarks would be highly dependent upon the type of toxic
effect observed. For a neurotoxicant, for example manganese,
this benchmark might be 50% inhibition of tyrosine hydroxylase
in the basal ganglia. Considering the great importance of this
enzyme in central control of neuromotor function, one might
apply a rather stringent divisor to the dose and consider that
a ten-fold less exposure is the highest ADI. But in the case
of a hepatotoxin (succinyl acetone), where the benchmark might
be 50% inhibition of the enzyme ALA dehydrase, knowledge of the
greater functional reserve of this system might permit a less
stringent division of the benchmark dose to estimate an ADI.
Thus, I am not certain that this approach really advances us
very much.
In the case of carcinogens, this approach would constitute
a significant change from the current methods of CAG. It
should be very carefully reviewed, and some estimation of its
impact, in a regulatory sense, should be undertaken by ECAO.
Some of us have had experience of similar revisionism in Dr.
Albert's 1981 proposals for carcinogens, which would have
raised the "ADI" for dioxin by about 2,000 times.
IV. Route-to-route conversion
EPA is interested in using data obtained from experiments using
one particular route of exposure in estimating health effects
of exposure through another route. This has clear utility and
would reduce the amount of experimentation needed before
regulation an airborne carcinogen, for instance, after a water
criteria document has been researched and published. It was
proposed that such route-to-route extrapolations could be
[3]
A-76
-------
acceptable under certain conditions: when the portal of entry
(lung; GI tract; skin) is not the target or critical organ;
when firstpass metabolism is not of major importance; and when
conditions of exposure (coupled with metabolism or activation)
are not such as to produce rapid changes in plasma levels in
intermittent exposures. Both Doctors Withey and O'Flaherty
presented interesting datasets indicating the pitfalls
presented by cases which violate these preconditions.
Consensus was genrally reached on this approach, although
some reservations were expressed on specific chemcials.
V. Multiple chemical assessment
EPA requires some method of integrating the combined risks of
exposure to multiple chemicals. A model was presented by
Hertzberg and Durkin related to dose additivity. There was a
lack of acceptance of this approach, particularly in its
fractional approach (i.e. those affected by the first exposure
would not be affected by the second, etc.). The model does not
appear relevant to the major questions related to multiple
chemical assessment: interaction of toxicants;
synergism/antagonism of specific agents; likelihood of
coexisting health effects in one individual (i.e.. tumor,
hepatotoxicity. nerve damage, immune dysfunction can all occur
in one animal exposed to dioxin; hepatotoxicity. CNS and
peripheral neuropathy, and hematoxicity can all occur in one
individual exposed to lead). I did not find this an approach
which was likely to be useful to ECAO.
In cases where one highly toxic substance can be identified
in a mixture (i.e.. dioxin), EPA may be better by basing
assessment on that substance alone. In case6 where substances
are identified and are known to affect different organs, then
each effect must be assessed and the overall result considered
within the framework of effects ranking (see discussion of
workshop topics below). In cases of unknowns, there is very
little to choose from in order to guide assessment. Dose
additivity is clearly insufficient, yet total separation of
each substance is not appropriate. This subject requires much
further discussion.
A-77
[4]
-------
NEW TOPICS
I. Structure-activity analysis
Two presentations were made on the use of chemical structure in
modelling an predicting toxicity. I found this discussion
baffling. It was admittedly unrelated to what is usually
considered the basis of structure-activity analysis—that is
the use of mechanistic hypotheses to generate "families" of
similarly acting agents. The methods described by Broderius
and Enslein appear to be solely empirical in that lists of
reactive chemicals are entered into the model and equations
derived from these lists. These lists work well on an internal
basis--that is, they predict the behavior of those chemicals
used to develop them. That is not terribly impressive. As
Broderius indicated, his models for bioaccumulation and fish
toxicity work best in describing compounds of relatively low
toxicity. The need for SAR is to warn us of the highly toxic
substances, not to rank low-concern chemcals. Comment was made
that "true" SAR of the type done by Kaufman at Hopkins is very
expensive. However, its success has great implications for
highly accurate and valuable predictive strategies. ECAO would
do well to support this type of SAR, rather than the type
presented at this meeting.
II. Reproductive toxicity
ECAO is concerned to develop methods of assessing risk of
reproductive effects of chemicals. This is an area of
tremendous public concern, and the Agency should overcome its
failure to propose standard methods of reproductive toxicity
testing under TSCA, as well as its lack of research on
reproductive effects in dumpsite studies.
Manson gave an elegant presentation on two types of
teratogens; a direct acting teratogen whose activity can be
predicted because of its special toxicity for cells in mitosis,
and an indirect teratogen whose less predictable effects appear
to be related to hormonal modulation of organ development. The
presentation indicated the need to develop wide-ranging and
sensitive indicators of adverse reproductive outcome. However,
by its focus, this discussion did not cover nonteratogic
reproductive toxins. I mentioned two other types likely to be
of great importance in ECAO's assessments: fetotoxins.
substances to which the fetus is more sensitive at lower dose
than adult organisms: and gonadal toxins, substances which
affect the sex organs, their physiological function, and the
[5]
A-78
-------
germ cells. No consensus was reached on this topic, and must
of the sense of the meeting was concern over the apparent
complexity of this area of toxicology. Further discussion is
greatly required, particularly since this field will be
important in dealing with issues of severity of effects and
specially sensitive subpopulations (see below).
III. Mathematical models
This presentation was also less than appropriate to the
complexity of real-world toxicological data. The model's
structure was highly linearized, and omitted any
Michaelis-Menten type kinetics which are highly relevant. It
is suggested that ECAO consider the model for procarcinogens
recently published by Hoel in Science as a more reasonable
approach to this type of modelling. In this topic as in
others, I am concerned about the lack of appreciation by ECAO
of the mechanisms and dose-respon6e relationships for
noncarcinogenic toxicants.
WORKSHOP TOPICS
I. Hypersusceptible subgroups
As indicated by Bingham. EPA i6 reconsidering its generic
approach to covering variations in human susceptibility by one
overall factor of 10. There was general consensus that this
range is likely to be insufficient to cover fully the range of
responses of major subgroups in the population, such as
children. Nisbet pointed out that children are at increased
risk by their greater rate of uptake (inhalation, ingestion);
their behavioral interactions with the environment; their
propensity to remain outdoors; their underdeveloped metabolism;
and their increased sensitivity of some critical target organs
(such as the brain). It was suggested that increasing the
factor might account for this, but as I pointed out. in some
instances different groups may actually present with a
different critical organ for toxic effect—e.g., the work of
Manson demonstrating the highly specific sensitivity of the
fetal heart for disturbances in T4/T3 metabolism.
Erdreich presented a good discussion of the data in
pharmacogenetics and proposed that analysis of this data might
be relevant for estimating the types of variance in response
[6]
A-79
-------
(e.g., log-normal or multi-modal). This was endorsed, and
additional sources of data were also suggested (such as the
Fels Institute longitudinal studies).
The paper of Hattis, restricted as it was to acute effects,
was not particularly helpful. Moreover, it was not accepted
that population variability should be incorporated into a
hazard index; the consensus of opinion was that this should be
factored into setting the ADI, which is then U6ed in
determining hazard indices and other assessment variables.
II. Use of exposure data in assessing
health risk
This workshop group focussed its discussion on a perceived
discontinuity between environmental monitoring/modelling and
epidemiology (or the monitoring/modelling of population factors
which determine exposure}. The group agreed that this lack of
communication, which affects both scientific disciplines and
the various offices of EPA, needs correction. Increased data
are generally needed, particularly in monitoring, and there
should be opportunity for feedback on second tier studies.
Further consideration of this topic should include
epidemiologists, particularly those who have studied exposure
as well as disease associated with that exposure.
On the other workshop topics, which I did not attend. I
have particular concerns about the ranking of severity of
effects. The table included in the premeeting materials was
not apparently much discussed. It is greatly biassed towards
pathological findings with hyperplasias and necrosis in all
cases ranked more severe than biochemical changes. Some
statements are so unqualified as to be uninterpretable--e.g..
what does "any decrease in reproductive capacity" include? The
difference between "death" and "pronounced life-shortening" is
not clear—how is life shortened except by death? Why is
teratogenicity without signs of maternal toxicity more serious
that teratogenicity with signs of maternal toxicity? It may be
the case that maternal effects are quite different (and
possibly less severe) from fetal effects. Some appreciation of
reversibility/ir- reversibility should be incorporated into
this table; also, possibly 6ome organ-specific
considerations—that is. it is probably always worse to damage
ova rather than 6perm, since sperm are replaceable while ova
are fixed in number during prenatal developments.
[7]
A-80
-------
OTHER COMMENTS
I would suggest consideration of the Monitoring and Assessment
Research Centre publications by ECAO in its further work on
these subjects. Moreover, more presentations of that they show
real toxicology data in some detail, so that we can better '
understand the utility of some of the models to these data.
For some other approaches to SAR, I suggest inviting Joyce
Kaufman or David Rodbard (NIH) to discuss real SAR approaches,
using computer models of agent-receptor interactions.
For another approach to interactions of toxicants, you
should consider the new work by Klug's group on carcinogenic
metals which act at catalytic sites in DNA (a new contribution
to the threshhold/nonthreshhold debate). Also. I enclose a
statement by the Dahlem Conference workgroup on mechanisms of
metal toxicity, which you may find useful. You could invite
Dr. Robert Willimas of Oxford to discuss his model approaches
to toxicity of metals.
OTHER TOPICS FOR DISCUSSION
1. Environmental epidemiology.
2. Quantal vs. continuous effect's of toxins.
3. Neurotoxicity.
4. Models of metabolic activation (Hoel).
5. Studies on waste streams as examples of multiple
chemical mixtures (see NY State Department of Health
studies on "Binghamton soot," attached).
6. Bioavailability.
A-81
[8]
-------
Health and Welfare
Canada
Health Protection
Branch
Sante et Bien-etre social
Canada
Direction generate de la
protection de la sante
Sir Frederick G. Banting
Research Centre,
Tunney's Pasture,
Ottawa, Ontario
KLA. 0L2
July 20, 1983
foui cue van reference
Dr. C. Cueto,
. . _ . Our me Notn f&6wnce
Project Director,
Dynamac Corporation,
The Dynanac Building,
11140 Rockville Pike,
Rockville, Maryland
U.S.A. 20852
Dear Dr. Cueto:
Attached are my post-meeting comments for the recent
methodology meeting held in Cincinnati July 12 and 13th last.
I thought that the meeting was, on the whole, fruitfull in
crystallizing seme of our thoughts on important areas. In others,
ws were less than definative.
If I were to be asked for suggestions on the "next step"
I vrould recartnend that we apply seme of our approaches to real
situations. There is probably enough data in the literature,
for example, to apply some route-to-route extrapolation procedures,
to assess ADI's for corplex mixtures (durtp site monitoring data)
and to do sane calculations an multiple route effects.
Above all, I must strongly recarmend that the presentations
and some of the discussion following, be published in a form which
can be referenced and cited. It would be a pity to go over old
ground repeatedly.
I very much enjoyed meeting with you and your staff. I trust
that all of my documentation, including copies of my presentations
and slide material, was in order. If you require additional help
with this please ask.
Toxicology Research Division,
Bureau of Chemical Safety.
JKW/mad
Attach.
A-82
-------
POST MEETING COMMENTS
Methodology Meeting, U.S. Environmental Protection Agency,
Cincinnati, Ohio, July 12-13th, 1983
1. Route to Route Extrapolation
(i) It would appear that our thoughts on the pharmacokinetic approach
have been consolidated to a point where, given the values of rate
coefficients, absorption coefficients and the exposure regimen,
extrapolation from inhalation to oral uptake, and the temporal
relationships for systemic levels, could be constructed. We agreed
that, in the case of the oral route, first pass effects are an
important aspect of the assessment but does not necessarily exclude
the application of quantitative route extrapolation procedures.
Sane special mechanisms, such as the deposition of aerosol and
particulate matter in the lung followed by uptake to the systemic
circulation; need to be considered as a factor which may preclude
route extrapolation.
(ii) The Stokinger-Woodward approach is crude and should be avoided
wherever possible. The S-W method, where gavel absorption coefficients
for both the oral and inhalation routes are available, will correctly
predict the relative amounts which are taken up into the systemic
circulation. As ny case 3 shews, relative rates of uptake and
elimination ray be uniirportant in determining the tenporal
relationship for systemic levels. In such cases, the S-W method
will be as good as any sophisticated pharmacokinetic methodology.
(iii) It would be interesting to have a list of those ccnpounds, which
are in the AMQC list, for which the S-W approach was used.
A-83
/2
-------
- 2 -
(iv) I should like to have an exposure scenario which might be
considered more realistic than the one I used to illustrate the
blood level-time curves in my cases 1, 2 and 3. You will recall
that I used a 10 hr. vapour phase exposure and either 4 or 20 oral
doses over a 10 hr. period. Perhaps a 24 hr. inhalation exposure
period might be more applicable. It would also help if different
pharmacokinetic models (ie. a 2 or 3 compartment) and specific
rates of uptake, distribution and elimination (the hybrid rate
coefficients) were used. Any feedback on this which would represent
a more realistic appraisal than those used, for illustrative purposes,
in my exanples will allcw me to feed my computer and generate seme
interesting figures.
2. Species Extrapolation
We were treated to a good presentation and seme lively discussion on
the biological rationale for dose as assessed on an animal body weight or
body surface area or sore power law of each of these. These relationships
are, however, empirical.
As I mentioned in the discussion, a much more rational approach has been
suggested by Dr. Jim Gillette in several publications (Gillette, 1976;
Gillette, 1977). I enclose a hard copy of both of these for your interest.
While we may not have all of the information required for this kind of
conversion it should be used as a beacon for further consideration.
I am particularly concerned about species with metabolic deficiencies.
I have reviewed this subject recently and can give you seme examples:
(i) Wide species differences in enzyme activity eg. a 50-fold range
for the metabolism of hexobarbital (Quinn et al., 1958).
A-84
— /3
-------
- 3 -
(ii) Different conjugative pathways for phenols. The pig and the rat
excrete the glucuronide and sulfate; sheep excrete phenols as the
phosphate (Kao et al., 1979).
(iii) Rats can not N-hydroxylate aliphatic amines, guinea pigs can not
N-hydroxylate aromatic amides nor produce mercapturic acids.
(Caldwell, 1981).
(iv) Cats do not produce the glucuronide of phenols or aromatic acids
and dogs find it difficult to acetylate primary amino groups or
to effect same kinds of acyclic hydroxylation (Caldwell, 1981).
(v) Small rodents tend to use glutathione for detoxification rather
than water and the ratio of glutathione transferase to epoxide
hydrase activities is mich higher in the rat and mouse than in
non-human primates and man. Thus the rat and mouse use valuable
protective materials as their first line of defence whereas man
and primates use glutathione only as a reserve after water and
epoxide hydrase have exercised the primary protective effect
(Pacifici et al., 1981).
(vi) A comparison of species as adequate metabolic models for man is
summarized in a review on this subject (Smith and Caldwell, 1976).
3. Multiple Exposures
I was a little concerned that the sum of intrinsic activities for a
series of caipounds in a mixture was accepted so readily. As I pointed out
in a recent appraisal (Withey, 1982) interactions can affect, quite
considerably, all aspects of pharmacokinetics. In particular, if I was asked
to assess the human health impact of a dump site which contained polycyclic
hydrocarbons, or other inducers of the mixed function oxidases, I should want
A-85
.. ./4
-------
- 4 -
to accept much lower estimates for the acceptable exposure to aromatic
halogen conpounds, like the PCB's, and other hepatotoxic and carcinogenic
compounds.
4. Exposure Data
I have little to say on this difficult issue. Clearly, the discussion
established that models, particularly for air, are difficult to construct.
It would seem to me that the figure 1. Information Flow and Tfethoctology
Use in Conducting Site-Specific Multichemical Risk Assessment, Showing Use
of Each Methodology Component Under Development by ECAO-Cincinnati can new
be reconstructed with a much larger "box" for Exposure Assessment. Someone
must address this topic and set the GOAL and type of METHODOLOGY USED. It
would also seem that there is a need to reorganize this diagram to illustrate
cross interaction at various stages. In conclusion, I think that the
methodology discussions have been hampered by a lack of site-specific
information.
J.R. Withey,
July 20, 1983
A-86
-------
REFERENCES
1. Caldwell, J. (1981) The current status of attempts to predict species
differences in drug metabolism,. Drug Me tab. Rev., 12_, 221.
2. Gillette, J.R. (1976) Application of pharmacokinetic principles in the
extrapolation of animal data to humans, Clin. Toxicol., 709.
3. Gillette, J.R. (1977) The phenomenon of species variations; problems
and opportunities in: Drug Metabolism from Microbes to Man, Parke,
D.V. and Smith R.L. eds., Taylor and Francis, London, 147.
id
4. Kao, J., Bridges, J.W. and Faulkner, J.K. (1979) Metabolism of ( 'C)
phenol by sheep, pig and rat, Xenobiotica, 9_, 141.
5. Pacifici, G.M., Boobis, A.R., Brodie, M.J., McManus, M.E. and Davies,
D.S. (1981) Tissue and species differences in enzymes of epoxide
metabolism, Xenobiotica, 11, 73.
6. Smith, R.L. and Caldwell, J. (1976) Drug metabolism in non-human primates
in: Drug Metabolism fran Microbes to Man, Parke, D.V. and Smith, R.L.,
eds., Taylor and Francis, London, 331.
7. Withey, J.R. (1982) Tbxicodynamics and Biotransformation in: Assessment
of Multichanical Contamination, Proceedings of an International Workshop,
National Acadary Press, Washington, D.C., 225.
A-87
-------
EPRI
Electric Power
Research Institute
September 7, 1983
Dr. Jerry F. Stara
Director, Environmental Criteria
and Assessment Office
U. S. Environmental Protection Agency
26 West Saint Clair Street
Cincinnati, OH 45268
Dear Jerry:
I am sorry to be late with my post-meeting memo.
Consensus Topic: A. Interspecies Conversion
The current approach is based on surface area differences.
The consensus appeared to be a request for greater flex-
ibility. In the presence of uncertainty, the most con-
servative approach (based on surface area ratios) should be
followed. There are cases, however, where alternative
methods (e.g. ratios of body weight) may be more suitable;
m such cases, they should be utilized.
I would personally urge that more use be made of pharmaco-
kinetic models where they can be demonstrated to fit the
available data.
Consensus Topic: B. Less than Lifetime Exposure
Although both options were clearly presented, there was no
clear consensus here with respect to toxicants. I believe
several more examples would help clarify the differences
between the two options as veil as indicate which performs
most consistently. Option 2 should consider alternative
dose-response curves to the log-log as well as consider
weighted curve-fitting procedures; which take into account
the quality and quantity of data used to derive the curve.
For carcinogens, there was a general consensus in favor of
option 3, although some modifications were suggested. The
Crump-Howe method could be simplified by applying it to some
prototype population distributions, and some allowance
should be made for those cases where the multi-stage model
does not appear to fit the data.
A-88
3412 Hiilview Avenue Post Office Box 10412 Palo Alto CA 94303 Telephone (415) 855-2000
Washington Office 1800 Massachusetts Ave. NW Suite 700, Washington DC 20036 (202) 872 9222
-------
Dr. Jerry F. Stara
September 7, 1983
Page 2
Consensus Topic: C. ADI's Based on Quantal, Continuous or
Graded Data
There seemed to be greater support for second option than
for the first although there still seemed to be some hold-
outs xn favor of the present NOAEL-safety factor approach.
Consensus Topic: D. Route-to-Route Conversion
There was general consensus that a flexible approach was the
best one. Lack of suitable data and the inconsistent per-
formance of existing models cannot lead to the recommenda-
tion that one approach be exclusively followed.
Consensus Topic: E. Multiple Chemical Assessment
There was greater support for the second option although a
suggestion was made that the first be undertaken as well in
an effort to bracket the true risk of multiple exposure. In
any case all of the factors associated with multiple expo-
sure should be identified.
Workshop Topic: Al. High Risk Subgroup
There was consensus that special attention need be given
high risk subgroups and that the safety factor of 10 ut-
ilized for population variability may not be sufficient to
protect all high risk subgroups. No consensus was reached,
however, about how to resolve the high risk subgroup problem.
Workshop Topic: Bl. Multiple Route Exposure
This group developed detailed guidelines with general agree-
ment about five conditions for summing exposures across all
routes.
1) dose-duration patterns should be identical across
routes.
2) System effects should be identical across routes.
3) there are no effects at portal of entry.
4) there are no first pass effects.
5) D/R/T known for at least one route in terms of internal
dose.
With lesser information, extrapolation estimates become more
tentative.
A-89
-------
Dr. Jerry F. Stara
September 7, 1983
Page 3
Workshop Topic: A2. Ranking Severity of Effects
There was support for the concept of ranking by severity of
effect but there was no agreement about the approach taken
to achieve the goal. More illustrative examples of alter-
native approaches may serve to clarify them and lead to more
definitive conclusions.
Workshop Topic: B2. Exposure Data
There was a consensus statement prepared about exposure
assessment. I believe, however, more detailed presentations
by Jim Falco of the EPA Exposure Assessment groups could
lead to even greater consensus about this topic.
New Topics; A. Approaches Using Structure-Activity Rela-
tionships
It would be useful to learn about additional approaches for
these models to compare the performance of alternative
techniques. I'm hesitant to use these approaches when other
(e.g., toxicologic or health) data are available however, in
the absence of any information these relationships appear to
be the only tool available.
B. Reproductive Effects
My interpretation of Jeanne Manson's talk was that these
effects are relatively rare and major attention should not
be given to these effects. I found Dr. Manson's presen-
tation to be clear, coherent, and convincing but the con-
clusion was contrary to my lay predisposition. For that
reason it might be useful to invite other well-known tera-
tologists to address the group at future meetings to ensure
that there is a reasonable consensus about these effects.
Perhaps someone familiar with the existing and developing
cl.inical dat£ bases in this area could be of help (i.e.
James Hanson, University of Iowa or a representative from
CDC) .
C. Use and Biological Justification of Mathematical Models
Dr. 0'Flaherty's presentation was clear and reinforced the
importance that should be addressed to first-pass meta-
bolism. Some provision for this effect should be incor-
porated into our work.
I hope these comments are helpful, if you would like any
further discussion of any of them, please feel free to
A-90
-------
Dr. Jerry F. Stara
September 7, 1983
Page 4
telephone me at (415) 855-2577.
Sincerely,
Ronald E. Wyzga, Sc.D.
Technical Manager
Environmental Risk Analysis
REW/cdl
cc: Cipriano Cueto
A-91
-------
SECTION 3
REFERENCES CITED IN MEMORANDA
Alarie, Y. 1981. Dose-Response Analysis In Animal Studies: Prediction of
Human Responses. Environ. Health Perspect. 42:9-13.
Bliss, C.I. 1939. The Toxicity of Poisons Applied Jointly. Ann. Appl.
Biol. 26:585-615.
Boxenbaum, H. 1982. Interspecies Scaling, Allometry, Physiological Time,
and the Ground Plan of Pharmacokinetics. J. Pharmacokinet. Blopharm.
10:201-207.
Caldwell, J. 1981. The Current Status of Attempts to Predict Species
Differences in Drug Metabolism. Drug Metab. Rev. 12:221.
DeCaprio, A.P., D.W. McMartin, J.B. Silkvorth, R. Rej, R. Pause, and L.S.
Kaminsky. 1983. Subchronlc Oral Toxicity in Guinea Pigs of Soot from
a Polychlorinated Blphenyl-Containing Transformer Fire. Toxicol.
Appl. Pharmacol. 68:308-322.
Gillette, J.R. 1976. Application of Pharmacokinetic Principles In the
Extrapolation of Animal Data to Human. Clin. Toxicol. 9:709.
Gillette, J.R. 1977. The Phenomenon of Species Variations; Problems and
Opportunities. In: Drug Metabolism from Microbes to Man, D.V. Parke
and R.L. Smith, eds., Taylor and Francis, London. 147 pp.
Kao, J., J.W. Bridges, and J.K. Faulkner. 1979. Metabolism of (*^C)
Phenol by Sheep, Pig and Rat. Xenobiotica. 9:141.
Nicholson, U.J. 1982. Proceedings of an International Symposium on Hazards
Related to Plastics and Synthetic Elastomers. Helsinki, Finland.
November 1982.
Nlsbet, I.C.T., and N. Karch. 1983. Chemical Hazards to Reproduction.
Noyes Data Corp.
Pacificl, G.M., A.R. Boobls, M.J. Brodle, M.E. McManus, and D.S. Davies.
1981. Tissue and Species Differences in Enzymes of Epoxide
Metabolism. Xenobiotica 11:73.
A-92
-------
Smith, R.L., and J. Cadwe11. 1976. Drug Metabolism in Non-Human Primates.
In: Drug Metabolism from Microbes to Man, D.V. Parke, and R.L. Smith,
eds. Taylor and Francis, London. 331 pp.
Withey, J.R. 1982. Toxicodynamics and Biotransformation: Assessment of
Mutichemical Contamination. In: Proceedings of an International
Workshop, National Academy Press, Washington, DC. 225 pp.
A-93
-------
SECTION 4
BACKGROUND REFERENCES
A-94
-------
BACKGROUND REFERENCES
Stara, J.F., M.L. Doursoti, and C.T. DeRosa. 1981. Water Quality Criteria:
Methodology and Applications. In: Conference Proceedings: Environ-
mental Risk Assessment. How New Regulations Will Affect the Utility
Industry, R.J. Hoch, ed., Sigma Research, Inc., Richland, Washington.
Crump, K.S., and R.B. Howe. 1983. The Multistage Model with a Time-
Dependent Dose Pattern: Applications to Carcinogenic Risk Assessment.
Prepared for U.S. EPA, ECAO-Cincinnati, OH. Contract No. 68-03-3111.
A-95
-------
Conference Proceedings: Environmental
Risk Assessment
How New Regulations Will Affect the
Utility Industry
EA-2064
Research Project 1316-6
EPRI Contract No. WS-80-134
Proceedings, October 1981
New Orleans, Louisiana
December 10-11, 1980
Prepared by
R. J. Hoch
SIGMA RESEARCH, INC.
2950 George Washington Way
Richland, Washington 99352
Prepared for
Electric Power Researcn Institute
3412 Hillview Avenue
Palo Alto. California 94304
EPRI Project Manager
R E. Wyzga
Environmental Risk and Issues Analysis Program
Energy Analysis and Environment Division
A-96
-------
Section 3
WATER QUALITY CRITERIA: METHODOLOGY AND APPLICATIONS
J. F. Stara et al.*
INTRODUCTION
In the last 20 years there has been an almost overwhelming proliferation of chemicals
in our environment. In 1977, for example, there were more than 60,000 conanercially
significant chemicals, and additional compounds were marketed at the rate of 500 to
700 per year U). These figures do not reflect the staggering diversity of compounds
developed but not produced in commercially significant quantities. They do illustrate
that water basins in the United States are being polluted by increasingly complex
municipal and industrial discharges (Figure 1). This contamination by point-source
pollution is predominant in the water basins of Northeast, Great Lakes, North Central
Regions, and Islands (Table 1).
Figure 1. Basins affected (in whole or in part) by
industrial discharges, 1977. Shaded areas show basin
areas contaminated by industrial discharge. (Source:
US EPA, national Mater Quality Inventory: 197? Report
to Congress, 1978)
* M. L. Dourson and C. T. DeRosa; Office of Environmental Criteria and Assessment,
US EPA, Cincinnati, OH.
A-97
-------
Table 1
POINT SOURCE POLLUTION BY REGION. 1977*
Percentage Affected by Type of Sourct**
Region
Number o1
Basins
Industrial
Municipal
Combined Sewei
Overflow
Northeast
40
65
95
60
Southeast
47
74
9)
17
Great Lakes
41
80
95
37
North Central
35
74
66
6
South Central
30
70
100
0
Southwest
22
23
64
0
North west
22
SS
73
¦\4
Islands
9
89
100
0
Total
246
72
89
20
'Source Reference 2
**ln whole or in pan
The ubiquitous nature of such pollution poses a difficult dilemma for public health
officials. A significant proportion of our population is being exposed to chemicals
that have yet to be assessed with respect to their potential health hazard. From the
standpoint of public health, possible biological effects include: acute, subacute,
or chronic toxicity; mutagenicity; teratogenicity; and carcinogenicity. Clearly, in
the absence of appropriate control technologies, the unrestrained chemical contami-
nation of the environment represents a significant threat to public health.
HAZARD EVALUATION METHODOLOGY
The issue of hazard evaluation or assessment deals specifically with potential routes
of exposure and associated health risks. Procedurally, such an assessment invokes
the basic toxicological concept of dose-response relationships; in environmental
terms, this translates as exposure versus risk. Although the Environmental Criteria
and Assessment Office-Cincinnati (ECAO-Cin) has been most concerned with pollutant
exposure through water, it is obvious that all routes of exposure should be con-
sidered in any assessment of hazard. For carcinogens, however, only incremental
risks associated with contaminants in ambient water were considered.
A further distinction must be made between voluntary and involuntary risks associated
with exposure. While society may routinely accomodate a defined risk level associ-
ated with a therapeutic drug for the treatment of a disease, it may reject any risk
A-98
-------
level associated with the presence of hazardous chemicals in drinking water. Occu-
pational exposure falls somewhere in the middle of this continuum. Workers in the
nuclear industry may, at tunes, accept a higher level of risk than the general pop-
ulation. On the other hand, the acceptance of enhanced risk levels is not entirely,
voluntary, and risks should be minimized whenever it is feasible to do so.
With the distinction between voluntary and involuntary hazards in mind, the precise
nature of the risk associated with exposure to a particular chemical agent can be
defined by using the biological endpoints discussed previously (e.g., toxicity,
carcinogenicity, etc.). It is the consensus of scientists and the National Academy
of Science (NASJ 'that these endpoints can be considered to be either threshold or
nonthreshold phenomena (_3). Biologically, threshold represents a no-effect level
explained by an organism's resistance or sum total of defense mechanisms in the face
of toxicological challenge. In contrast, chemical carcinogens are considered to be
nonthreshold agents, since a single genotoxic molecule can be assumed to interact
with the cell's DNA and, thereby, result in a malignant growth. While not all car-
cinogens are genotoxic, epigenetic carcinogens were treated conservatively using the
nonthreshold hypothesis since sufficient scientific data are not yet available to
resolve this issue.
The problems presented by chemical contamination of water resources are not unique
to the United States. As .a consequence, a growing number of countries have enacted
legislation to regulate production, use, and release of chemicals (Table 2). The
international flavor of the issue is further amplified by the fact that the World
Health Organization (WHO) has proclaimed the 80's as the "Decade of Clean Water."
Within the United States, the responsibility for regulatory measures and the devel-
opment of guidelines is assigned to a number of agencies (Table 3).
The initial quandary to be confronted by such organizations and agencies is the
bewildering ocean of chemicals that represent a potential hazard to human health.
As a result, the first issue of concern is the relative priority of chemicals for
hazard evaluation. Criteria for determining a specific chemical's priority for
subsequent evaluation include:
• Extent of production and industrial waste;
• Use pattern;
• Persistence;
• Dispersion tendency;
A-99
-------
Table 2
COUNTRIES AND ORGANIZATIONS THAT REGULATE CHEMICAL SUBSTANCES
Country
c
0
je
1
-i
Date
Canada
Environmental Contsminvm Act
1B74
France
Control of Chemical Products Act
1977
Federal Republic of Germany
Chemicals Act
Pending
Japan
Chemical Substances Control Act
1673
Sweden
Act on Products Hazardous to Man and the Environment
1973
Switzerland
Law of Trade m Toxic Substances
1069*
United States
Toxic Substances Control Act
1876
Denmark
Toxic Chemicals Act
Pending
European Communities
Directive for the Sixth Modification of the Council Direc-
tive of 27 June 1967 on the Approximation of the Laws
of Member States Relating to the Classification. Packag-
ing, and Labeling of Dangerous Substances.
1979
'Effective 1972
Table 3
UNITED STATES REGULATOR Y AGENCIES
Agency
Established
Authority
PHS
1798
Public Health Service Act of 1944
USDA
1852
Act of Congress
FDA
1907
Food and Drug Act
osha
1970
Occupational Safety and Health Act
EPA
1970
Executive Order-
Clean Air Act
Clean Water Act
Toxic Subnancet Control Aci
CPSA
1972
Consumer Product Safety Act
NIOSH
1973
National Institute of Occupational Safety and Health
Established within PHS by Secretary of Health,
Education, and Welfare
A-100
-------
• Biotic and abiotic transformation; and
• Human toxicity {most important).
Secondly, regulators must determine the best approach for the evaluation of human
health hazard based upon a particular toxicological testing strategy. This often
encompasses a third problem that relates not only to the extrapolation of animal data
to man but also from short-tern animal studies to lifetime exposure for human popu-
lations.
The final issue is cost. At present, an NCI bioassay can cost as much as $500,000;
the required expenditure to test the 700 new chemicals of commercial significance
produced each year would be $350 million. Even if government and industrial expend-
itures were properly integrated to minimize duplication of effort, the cost of an
advisable testing protocol could not be met at current spending levels. As a result,
the US population uses and is exposed to untested, or inadequately tested, chemicals
of unknown toxic potential. Indirect cost considerations include the costs of regu-
lation reflected by reduced industrial productivity and the costs to society in terms
of hunan health in the absence of reasonable controls.
In promulgating controls and standards, government agencies have three distinct op-
tions. If sufficient evidence exists that a particular substanoe is a hazard to
public health, it can be banned—as in the case of Xepone. Alternatively, the use of
a hazardous chemical such as DDT can be limited. Finally, the limited release of
certain materials in industrial effluents can be permitted. This option is usually
pursued in the case of complex mixtures such as auto exhaust, the release of which is
limited by means of catalytic converters. The second and third options underscore
the fact that regulation does not necessarily stipulate total elimination; instead,
it may involve the adjustment of chemical levels in the environment as a function of
need. This, of course, is in accordance with the philosophy of risk-benefit analysis.
In many instances, regulatory agencies simply recounend that the levels or uses of
hazardous compounds be limited.
METHODOLOGY FOR CRITERIA DERIVATION
Either directly or indirectly, all of the issues discussed previously were involved
in the development of the 65 Consent Decree Hater Quality Criteria Documents (Table
4). Procedurally, the documentation process entailed an in-depth world literature
review. The resulting information was used in generating four chapters that relate
directly to mammalian toxicology and human health effects. Critical peer review by
A-101
-------
250-300 scientists representing the governmental, academic, and industrial commun-
ities ensured the quality and objectivity of these documents.
Table 4
THE 65 CONSENT DECREE WATER CRITERIA DOCUMENTS
Acenaphthene
Cyanide
Lead
Acrolein
DDT
Mercury
Acrylonitrile
Dichlorobenzene
Napthalene
Aldnn/DieMrin
Dichlorotwnzidine
Nickel
Antimony
Diehloroethylenes
Nitrobenzene
Arsenic
2.4 -Dichloropheno I
Nitrophenolt
Aibettot
Dichloropropanes/penes
Nitrosemines
Benzene
2,4-Dimethyl phenol
PAH's
Benzidine
2,4-Dinitrototuene
PCB't
Beryllium
Dioxini (TCDDI
Pentachloroohenol
Cadmium
Diphenylhydrazirve
Phenol
Carton Tetrachloride
Endosuffan
Phthalate Esters
QM oi dane
Endnn
Selenium
Chlorinated Benzenej
Ethylbenzene
Silver
Chlorinated Ethanes
Fluoranthene
Tttrachloroethyicnes
Chlorinated Napthalene
Haloethen
Thallium
Chlorinated Phenol*
Halomathanes
Toluene
Chloroalkyl Ethers
Heptachlor
ToMphene
Chloroform
He*aehlorobut»diene (HCBD)
Triehloroethylenes
2-Chlorophenol
Hexachlorocydohezane
Vinyl Chloride
Chromium
Hexachlorocydopentadiene
Zinc
Copper
IfoOhorone
The first chapter dealt with sources and levels of exposure as well as estimates of
total exposure and body burden. In most cases, levels of pollutants in ambient
waters were lower than recommended criteria. In some cases, however, pollutant
levels in ambient waters were found to be greater than the criteria due either to
natural background or anthropogenic releases, TOeoretically, criteria are levels of
pollutants protective of human health. Economic, social, and political considera-
tions must also be addressed in the promulgation of standards.
Die second chapter focused upon relevant pharmacokinetic parameters. Whenever
sufficient data were available, they were used in the derivation of criteria. Addi-
tionally, pharmacokinetic information provided guidance in the identification of
target organs, physiological half-tunes, and areas for future research.
A-102
-------
All toxic effects associated with acute, subchronic, and chronic exposure were
reviewed in the third chapter. These effects include: toxicity (acute, subacute,
and chronic), teratogenicity and other reproductive effects, mutagenicity, carcino-
genicity, and synergism. The most pertinent toxic effects were discussed in relation
to criteria derivation in chapter four.
The fourth chapter of the human health effects section specifically addressed the
derivation of criteria. Ambient water quality criteria are based on three possible
types of biological endpoints: carcinogenicity, toxicity (i.e., all adverse effects
other than cancer), and organoleptic effects. As discussed above, carcinogenicity is
regarded as a nonthreshold phenomenon. Therefore, safe or no-effect levels for
carcinogens cannot be established because any dose must be assumed to elicit a
finite response.
For compounds that do not manifest any apparent carcinogenicity, the threshold as-
sumption is used in deriving a criterion. This assumption is based on the premise
that a physiological reserve exists within the organism and must be depleted before
clinical disease ensues.
In some instances, criteria are based on organoleptic characteristics such as thres-
holds for taste or odor. These criteria are established when insufficient informa-
tion is available on toxicity or when the criterion based on organoleptic effects is
lower than the level calculated from toxicologic data. It should be recognized that
criteria based solely on organoleptic effects do not necessarily represent approxi-
mations of acceptable risk levels for human health (£).
Several ambient water quality criteria documents deal with classes of compounds that
include chemicals exhibiting varying degrees of structural similarity. Prediction of
biological effects based solely on structural parameters is difficult; therefore, the
derivation of compound-specific criteria is preferable to a class criterion. For
members of some chemical classes, however, compound-specific criteria cannot be
derived based on available data on one member of a class.
Cancer
After the decision has been made that a compound has the potential to cause cancer in
humans and that data exist to permit the derivation of a criterion, the water con-
centration associated with a lifetime carcinogenic risk of 10"5 is estimated. The
data used for quantitative estimates are of two types:
A-103
-------
• Lifetime animal studies, and
• Human studies where excess cancer risk has been associated with
exposure to the agent.
The procedure for deriving water concentration from animal studies is the improved
multi-stage model developed by Crump (5) and certain techniques developed by the US
EPA (45 FR 79350).
If human epidemiologic data and sufficiently valid exposure information are available
for the compound, the data are analyzed by an alternate procedure to give an estimate
of the linear dependence of cancer rates based upon the calculated lifetime average
dose. If the epidemiology data show no carcinogenic effect when positive animal
evidence is available, it is assumed that a risk exists but is smaller than could
have been observed in the epidemiologic study. An upper limit of the cancer inci-
dence is calculated, assuming that the true incidence is ;just below the level of
detection in the cohort studies. In human studies, the response is measured in terms
of excess risk of the exposed cohort of individuals compared to the control group.
In the analysis of these data, it is assumed that the excess risk is proportional to
the lifetime average exposure and that it is the same for all ages (45 FR 79350).
Both of these procedures yield slopes termed qi*(A) or B , respectively. Since
n
qj*(A) is derived from animal studies, it must be adjusted to yield an equivalent
human slope, qiMH), by the following equation:
.(H) . fa 1 * (A)) [ (mg/kg^d) ~1 ] >/ 70_k£
qi 1 ' (le/Le) (Le/L)J \ w(kg) u'
where:
qi*(A) » the upper 95% estimate of the linear component of the slope
(potency factor) estimated from all the animal data;
le ¦ the length of exposure;
Le ¦ the length of the experiment;
L ¦ the lifespan of the animal; and
w ¦ average weight of the animal.
The (Le/L)9 factor penalizes the calculation of qi*(H) from less than lifetime
animal studies. For example, if a study lasted only one-half the normal lifespan of
the animal, then, (1/2)® ¦ 1/8. This value causes an eightfold increase in the value
A-104
-------
of qi*(H) over what it would be if Le and L were equal.* The rationale for the use
of this factor is documented elsewhere (45 FR 79350) but is based on the concept
established by Druckrey (£) that time is more important than dose rate in a carcino-
genesis experiment. The cube root of the body weight ratios is a further refinement
of the qi*(A) potency factor and is based on the suggestion of Mantel and Schneiderman
(7) that it represents an equivalent dose among species.
After the slopes describing carcinogenic potency in humans have been calculated, the
intake (I) associated with a specific risk (usually 10~5 or 1 in 100,000) over a
human lifetime is determined:
Iftng/d) - 70k^l0-S) or_r0*L(10^) • <*>
1 9/ lqj«(H>] [(mgAg/d) '] BH(mgAg/d) *
For this calculation, the average weight of a man is assumed to be 70 kg. The ambi-
ent water quality criterion is a straightforward calculation:
,f) 1 (mg/d)
tm9/ ' " (2 t/d) + {[0.0065 kg/dj [BCF(tAg)]i '
The assumed average daily consumption of water and fish for a 70-kg man is 2 liters
per day and 0.0065 kilograms per day, respectively. BCF is the bioconcentration
factor of the chemical in liters per kilogram.
Toxicity
Zn developing guidelines for deriving criteria based on noncarcinogenic responses,
five types of response levels are considered!
• NOEL - No-Observed-Effect Level,
• NOAEL - No-Observed-Adverse-Effect Level,
• LOEL - Lowest-Observed-Effect Level,
• LOAE1 - Lowest-Observed-Adverse-Effect Level,
j
• FEL - Frank-Effect Level.
* Higher values of qi*(H) are associated with lower and, therefore, more protective
criteria.
A-105
-------
Adverse effects are defined as any effects that result in functional impairment and/
or pathological lesions that may affect the performance of the whole organism or that
reduce an organism's ability to respond to an additional challenge. Prank effects
are defined as overt or gross adverse effects (severe convulsions, lethality, etc.).
These concepts are illustrated in Figure 2. They have received nruch attention be-
cause they represent landmarks that help to define the threshold region in specific
experiments. Thus, if an experiment yields a NOEL, a NOAEL, a LOAEL, and a clearly
defined FEL in relatively closely spaced doses, the threshold region has been rela-
tively well defined. Such data are very useful in deriving a criterion. On the
other hand, a clearly defined FEL is of little use in establishing criteria when it
stands alone because such a level gives no indication of how far removed it is from
the threshold region. Similarly, a free-standing NOEL has little utility because
there is no indication of its proximity to the threshold region.
100
80
A SLIGHT BODY WEIGHT
DECREASE
B LIVER NECROSIS
C MORTALITY
DOSE (ARBITRARY UNITS)
Figure 2. Response levels considered in defining threshold regions in
toxicity experiments. Doses associated with these levels are as follow:
3 - NOEL, NOAEL; 4 - LOEL, NOAEL: 5 - NOAEL (Highest); 7 - LOAEL; 10 -
FEL; 20 - FEL.
A-106
-------
Based on the above dose-response classification system, the following guidelines for
deriving criteria from toxicity data have been adopted.
• A free-standing FEL is unsuitable for the derivation of criteria.
• A free-standing NOEL is unsuitable for derivation of criteria.
If multiple NOELs are available without additional data on LOEL's,
NOAEL's, or LOAEL's, the highest NOEL should be used to derive a
criterion.
• A NOAEL, LOEL, or LOAEL can be suitable for criteria derivation.
A well-defined NOAEL from a chronic (at least 90-day) study can
be used directly, applying the appropriate uncertainty factor.
For a LOEL, a judgment must be made as to whether it actually
corresponds to a NOAEL or a LOAEL. Zn the case of a LOAEL, an
additional uncertainty factor is applied; the magnitude of the
additional uncertainty factor is judgmental and should lie in the
range of 1 to 10. Caution must be exercised not to substitute
Frank-Effeet Levels for lowest-Observed-Adverse-Effect levels.
• If—for reasonably closely spaced doses—only a NOEL and a LOAEL
of equal quality are available, the appropriate uncertainty factor
is applied to the NOEL.
In using this approach, the selection and justification of uncertainty factors are
critical. The basic definition and guidelines for using uncertainty factors has been
given by the National Academy of Science (3_> • "Safety factor" or "uncertainty factor"
is defined as a number that reflects the degree or amount of uncertainty that must be
considered when experimental data in animals are extrapolated to man. When the
quality and quantity of experimental data are satisfactory, a low uncertainty factor
is used; when data are judged to be inadequate or equivocal, a larger uncertainty
factor is used. In those cases where the data do not completely fulfill the con-
ditions for one category—or appear to be intermediate between two categories—an
intermediate uncertainty factor is used. Such an intermediate uncertainty factor can
be developed based on a logarithmic scale (e.g., 33 being halfway between 10 and 100
on a logarithmic scale).
The actual procedure for calculating the ambient water quality criteria based on
toxicity data is quite simple. The chosen effect level in milligrams per kilogram
per day (mg/kg/d) from the animal study is divided by the appropriate uncertainty
factor. This is multiplied by 70 kg (the standard weight of a man) to yield an ac-
ceptable daily intake (ADD; then, the criterion (C) is calculated by an equation
similar to Equation (3) with the standard assumptions of water and fish consumption
as previously mentioned.
A-107
-------
C(mgA) = (2 £/d) + { [0.0065 kg/d] [BCF(iAg) 1 i
(4)
Organoleptic
Criteria based upon organoleptic effects are intended to protect against adverse
appearance, taste, and/or odor qualities in ambient water. While these criteria do
not necessarily represent safe levels of exposure for man because they are not based
on toxicity data, they may provide guidance to regulatory agencies in the promulga-
tion of standards. In the event that the organoleptic effect level is below the
level responsible for health effects, they can be used as the basis for recossnending
an ambient water quality criterion.
EXAMPLES
Cancer
The study of Preussman et al. (£) is an example of the basis of a cancer-based
criterion. Sprague-Dawley rats of both sexes were administered N-nitrosopyrrolidme
at one of several drinking water concentrations equivalent to 0, 0.3, 1.0, 3.0, or 10
mgAg/body weight/day throughout life. The corresponding incidences of various
malignant tumors were 6, 12, 20, 32, and 11%, respectively.
An estimated value of 0.364 (mgAg/d)-1 for qi*(A) was derived from these data using
the multi-stage model as previously described (5). The equivalent human potency
factor is then calculated by Equation (1):
0.364 (pq/kq/d)
(728 d/728 d)(728 d/728 d)
1
- 2.13 (mgAg/d)"1
The daily intake associated with a 10"5 cancer risk is determined by Equation (2):
j , 70 kg(10~5)
2.13 (mg Ag/d)"'
- 0.329 yg/d .
The ambient water quality criterion is found by Equation (3):
A-108
-------
0.329 Uq/d
(2 1/d) + 1(0.0065 kg/d)(0.055 l/kg))
™ 164 ng/£
Toxicity
The study of MacKenzie, et al. (9) is an example of the basis for a toxicity-based
criterion. Sprague-Dawley albino rats of both sexes were given chromium VI as po-
tassium chromate at one of several drinking water concentrations (0, 0.45, 2.2, 4.5,
7.7, 11, or 25 mg/£) for one year. No effects were noted at the six lowest doses.
The highest dose.(25 mq/S.) caused a decreased water intake.
For purposes of criterion derivation, the highest dose can be considered a well-
defined NOAEL (the lower doses are NOEL's). Furthermore, it is considered that the
25 mg/£ concentration represents exposure from chromium alone. An ADI can be calcu-
lated from this NQAEL as follows:
*9<" t/in -J-5 "W* .
in . >9' ¦ m/i ¦
Where 0.035 £/d is the assumed average daily water consumption and 0.350 kg is the
assumed average body weight of rats in the study, 1,000 represents the uncertainty
factor employed per NAS guidelines (3) and 70 is the assumed average weight of a man.
A protective level is calculated using Equation (4):
C ¦ mg/u m e3 .£
(2 Z/d) + [(0.0065 kg/d)(16 l/*q))
Based on these calculations, the US EPA recansnended the previous standard for total
chromium, 50 Vq/l (10), as the ambient water quality criterion for chromium VI. The
above calculations were judged supportive of the standard but insufficient to warrant
changing it.
SUMMARY
From the foregoing discussion, it is clear that agencies involved in environmental
regulation are confronted with a complex constellation of issues. The magnitude of
A-109
-------
risks posed by the three biological endpoints that were used as a basis for criteria
derivation (i.e., carcinogenicity, chronic toxicity, and organoleptic effects) are
determined ultimately by the levels of the compounds in the environment. Similarly,
the compartment(s) of the ecosystem in which a contaminant resides will often deter-
mine which agency or agencies will bear primary responsibility for regulation.
From a practical point of view, the greatest difficulty to be dealt with by regula-
tory agencies is the inadequacy of the toxicological data base. Long-term studies in
experimental animals should constitute a primary focus of future research for hazard-
ous chemical pollutants. Without such basic data, the scientific coronunity will be
hard pressed to make firm or defensible reconsnendations for the protection of human
health.
This need is, of course, complicated by economic considerations. The cost of aug-
menting the chronic toxicity data base for chemicals included in the 65 Consent
Decree documents is at least 50 million dollars. Zf all other commercially signif-
icant chemicals were considered, the cost would be much more. Consequently, regula-
tors, in concert with the scientific community, must develop a rational scheme for
determining research priorities. In toxicology, this need is being partially ad-
dressed by the proliferation of screening tests for carcinogenicity. Although these
tests certainly have merit, they have yet to be extensively validated.
Until toxicologists can provide reliable and economic screening procedures, we must—
to a great extent—rely upon environmental chemists. Chemical monitoring data pro-
vide reasonably precise estimates of chemical prevalence and current levels of human
exposure and should be considered in assessing research needs. While prevalence and
persistence do not necessarily indicate hazard, it is clear that biological testing
should emphasize hazard assessment for those chemicals to which the human population
is most extensively exposed.
REFERENCES
1. US EPA. "TSCA Candidate List of Chemical Substances." Office of Toxic Sub-
stances, Washington, DC, 1977.
2. US EPA. "In-depth Studies on Health and Environmental Impacts of Selected
Water Pollutants." Contract No. 66-01-4646. US Environmental Protection
Agency, Washington, DC, 1978.
3. Drinking Water and Health. National Academy of Science, Washington, DC, 1977.
A-110
-------
4. J. F. Stara et al. "Human Health Hazards Associated With Chemical Contamina-
tion of Aquatic Environment," Environmental Health Perspectives, Vol. 34,
pp. 145-158, 1980.
5. K. S. Crump. "An Improved Procedure for Low-Dose Carcinogenic Risk Assessment
From Animal Data." Jour. Environ. Pathol. Toxicol. (In press), 1980.
6. H. Druckrey. "Quantitative Aspects of Chemical Carcinogenesis." Potential
Carcinogenic Hazards from Drugs, Evaluation of Risks, R. Turhaut (ed.)f UICC
Monograph Series, Vol. 7. Springer-Verlag, New York, 1967.
7. N. Mantel and M. A. Schneiderman. "Estimating Safe Levels, a Hazardous Under-
taking. " Can. Res. 35, p. 1379, 1975.
8. R. Preussman et al. "Carcinogenicity of N-nitrosopyrrolidine: Dose-Response
Study in Rats." Z. KrebsForsch, 90, p. 161, 1977.
9. R. D. MacKenzie et al. "Chronic Toxicity Studies. II. Hexavalent and Triva-
lent Chromium Administered in Drinking Water to Rats." AMA Arch. Ind. Health.
18, p. 232, 1958.
10. US EPA. "Quality Criteria for Water." Washington, DC, 1976.
QUESTIONS AND ANSWERS
Q: As I understand it, you've said that in developing standards the states will
utilize the criteria as guidance input. Has EPA done any work to provide mo-
del analyses for the states to set regulations, or will this be left up to tne
states entirely?
A: I think Joe Cotruvo could probably answer the question better than I; however,
1 know that the federal representatives from the Water Quality Office and the
state people plan to get together. Fortunately, they asked that we be involved
along with the Cancer Assessment Group. Die issue of state standards will be
the subject of detailed review at these meetings.
Q: What about the role of state advisory groups? 1 alluded to them earlier
this morning. Can you tell us what you think some of the major topics for
analysis by these groups will be and how you see them helping the states in
setting regulations?
A: The only contacts we had were with a planning group in the state of Michigan.
I am not aware of how many preliminary meetings the Office of Water Quality
held with other state representatives.
Q: What is the basis of the National Academy of Science's safety factors and its
relationship to the acceptable daily intake?
A— 111
-------
A: The recomnended safety factors, or to use a better tern, uncertainty factors,
were published in Drinking Water and Health by the National Academy of Science
in 1977. The monograph also discussed the reasons why the safety factors are
to be used. In my talk, I have presented a summary of the recosmended approach.
The HAS recommends the use of uncertainty (safety factors) of 10, 100 or 1000
to provide for the uncertainty inherent in reported epidemiological and toxico-
logical data. This procedure was recommended by a Committee of the National
Academy of Sciences and our scientific groups considered this to be a useful
approach. Accordingly, we have adopted their use in our Methodology for the
Water Quality Criteria Documents. The Acceptable Daily Intake for a chemical
is then calculated using the formula I have presented which includes the appli-
cation of the appropriate uncertainty factor.
Qt As for state standards, what do you envision as the number of standards that
would have to be set in terms of the priority pollutants? Will the states have
to set numerical st&ndards for each one of these, for a certain percentage, or
what?
A: To my knowledge, the present official position is that US EPA will recommend
a list of pollutants to the states for which standards should be considered.
Some of the pollutants may not be considered because the data base is not as
strong as we would like it to be. In addition, using the linear model, some
of the criteria levels may be very low. Yet, as Dr. Albert said, we have used
the nonthreshold linear approach for carcinogenic chemicals. To explain this
further, when this model is used we estimate an incremental risk to the popu-
lation due to intake from water and contaminated fish over a 70-year period.
There is a possibility that either natural or anthropologic background levels
of the pollutant measured in ambient water may in some instances have been
higher than the calculated criteria level.
I must emphasize that the derived criteria correspond to levels of pollutants
in bodies of water which represent an estimate of incremental risk for carcino-
gens and "no effect levels" for noncarcinogens, and which may—at times—be
lower than reported concentrations of the pollutants in some areas. The criteria
are intended to be protective of human health without regard to present levels
in water. It should be our goal to reduce pollution levels accordingly, even
though—at the present time—we are unable to accomplish it. This is why the
criterion is only one of the factors that must be considered in setting the
state standards.
A-U2
-------
I still have one problem that I think is going to be a consistent problem:
How do we know what is a good animal study and what's a marginal animal
study? What is an intermediate study? When do we divide by 100, or 1,000, or
10,000? Ifrese criteria are not explicit and have never really been addressed.
I would think that we would have to do something about this in the future.
A: Let me just say that we had 8 or 10 expert scientists to review each document.
Three or four were scientists who had published extensively on that particular
chemical and can be called experts. TVo or three had expertise in the general
areas of epidemiology or toxicology; and one or two had expertise in cancer
research, tfhat is how our review committees were usually formed. In addition
to this, there were several government scientists who had an intimate knowledge
of the pollutant and of the related regulatory approach. The Committee made
the best possible judgments, yet 1 agree that there were—at tunes—unanswered
questions, and yours is one of them. Often scientific judgment is important
along with the data in resolving these issues. This is why I am satisfied that
our group will be involved in the discussions concerning setting of standards.
We know which data are satisfactory and where there are some gaps in knowledge.
Q: To carry Tony Colucci's comment one step further, tell us how you cope at
present with conflicting results from different studies? What method do you
have to resolve this? For example, suppose you have animal studies with con-
flicting results; one lab shows a positive effect and another lab shows a
negative effect. What do you do?
A: That's another good question, and 2 will give a very simple answer. If two
studies are comparable in quality and only one showed a positive effect, we
would use the positive study because, in studies with 20-100 animals, you cannot
dismiss the statistical probability that the negative data are equivocal. For
example, in the case of trichloroethylene, one NCI bioassay study demonstrated
carcinogenic effects in mice. Yet the studies of Van Duuren and Maltom are
reported to be negative. Can a regulatory agency contradict the NCI data? Can
we brush the NCI study under the rug—especially if it deals with a carcinogenic
response? Zn case of toxicity data, there are often discrepancies among various
studies; however, since these are threshold agents, the judgments concerning
conflicting effect data may be resolved more easily. But in the case of carcin-
ogens, we must be extremely careful.
A-113
-------
Q: What is the state of the data that you need before you consider it?
Must it appear, for example, in peer review journals?
A: Yes. At tunes, if we need further details, we may call the researcher either
to confirm the published data with additional information or, in case of a
continuing research project, we may be interested if the new data clarified the
gaps in information. Normally, however, we would use peer reviewed journal
articles almost exclusively.
Q: Do you always use the most sensitive response with noncarcinogens?
A: No.
Q: How do you decide?
A: Sometimes we may use a LOEL instead of a NOAEL. In other words, we use the
lowest-observable-effect-level if the study has a greater degree of reliability
than the study that suggested a NOAEL. We do not like to use a NOEL because
this factor usually represents an undefined number. In contrast, a NOAEL is
usually rather well defined.
Q: I was referring to qualitatively different kinds of responses. For example, if
you have an enzyme impairment as the most sensitive, there's a real question as
to whether or not that had any toxicological consequences.
A: In our methodology, we have described the types of effect data to be used;
they are based on functional effects and/or pathological lesion effects. At
times, biochemical effect data were used but only those that are of generally
recognized significance.
Q: Why doesn't EPA require peer review in research they do themselves?
Sometimes they issue documents that have no peer review at all.
A: This is possible. However, the scientific documents issued by ECAO-CIN,
vhich is my group, are all subjected to peer review.
A-1M
-------
-1
The Multistage Model With a Time-Dependent Dose Pattern:
Applications to Carcinogenic Risk Assessment*
(Abbreviated Title: Cancer Model with Time-Dependent Dosing)
2 3
Kenny S. Crump and Richard B. Howe
Address all correspondence to: Kenny S. Crump
Science Research Systems, Inc.
1201 Gaines Street
Ruston, Louisiana 71270
* This research was supported by EPA Contract 68-01-3111. It has not
been reviewed by EPA.
2
Science Research Systems, Inc., 1201 Gaines Street, Ruston, LA 71270
3
Dept. of Mathematics and Statistics, Louisiana Tech University,
Ruston, LA 71270
A-115
-------
-2-
Key Words: cancer, risk assessment, multistage model, exposure pattern.
A cancer risk assessment methodology based upon the Armitage-Doll
multistage model of cancer is applied to animal bioassay data. The
method utilizes the exact time-dependent dose pattern used 1n a bioassay
rather than some single measure of dose such as average dose rate or
cumulative dose. The methodology can be used to predict risks from
arbitrary exposure patterns including, for example, intermittent expo-
sure and short-term exposure occurring at an arbitrary age. The method-
ology is illustrated by applying it to a National Cancer Institute
bioassay of ethylene dibromide in which dose rates were modified several
times during the course of the experiment.
A-116
-------
-3-
I. Introduction
In the Armitage and Doll^ multistage model of cancer, it is
assumed that a cell line goes through k distinct stages in its pro-
gression to becoming cancerous. The rate at which it progresses through
the ith stage is assumed to be a constant Different cell lines are
assumed to compete independently in producing tumors. This simple model
LI
predicts that cancer incidence will increase as (age) , which agrees
with the observation that the age-specific incidence rates of many human
cancers—particularly carcinomas in organs other than sex organs-
increase as (age)* where x ranges between 3 and 6.^"^
The Armitage and Doll model has been extended to include the effect
of exposure to a carcinogen by assuming that the transition rate at
which a cell goes through each stage is linearly related to the dose
rate, i.e.t that
Xi (1)
Ml
where d is the dose rate of a continuously applied carcinogen/ ' Here
is the background transition rate in the absence of an applied dose,
and ^ represents the increase in the transition rate per unit dose. An
equivalent assumption is that the residence time of a cell in each stage
has an exponential distribution with mean 1/^. With this formulation
the probability of a cancer by age t is approximately of the form
. k
P(d,t) = 1 - exp[-ct ff (rfi +Pid)] (2)
i«l 1 1
where c is a constant that does not involve d, t, or any of the«'s or
A-117
-------
-4-
^'s.^ For a fixed age t, the exponent in this expression is a poly-
nomial 1n the dose rate d. This polynomial form prompted Guess and
(S)
Crumpv ' to recommend a model for carcinogenic risk assessment from
animal data of the form
P(d) = 1 - exp(-qQ - q^d-.. ,-qj(dk) (3)
where qi > 0. Here P(d) represents the probability of cancer by a fixed
time—typically the time at which the experiment 1s terminated. This
model, along with related statistical methods,has been used by
the EPA and other regulatory agencies to assess low dose cancer
risks.Although expression (3) has been called "multistage
model," it actually 1s more general than (2) with t fixed, because the
exponent in (3) contains polynomials not predicted by (2).
(13)
Whittemore and Keller* ' developed expressions for the Armitage-
Doll model which apply when the dose rate is not constant throughout
(141
life. Day and 8rownv ' extended this work and developed expressions
for the cancer patterns predicted by the multistage model when exposure
occurs early or late in life and when different stages in the carcino-
genic process are affected by the exposure. Comparing the cancer
patterns predicted by these expressions with experimental and epidemio-
logical data, they concluded that some of the known carcinogens appeared
to affect early stages ("early stage carcinogens") while others affected
late stages ("late stage carcinogens").
The purpose of the present paper is to develop further the model
studied by Whittemore and Keller and by Day and Brown, and to consider
its use in predicting carcinogenic responses to different dose patterns.
In particular, methods are developed and illustrated for applying the
A-118
-------
-5-
model to data from animal bioassays in which the administered dose rate
changes during the course of the experiment, and for developing esti-
mates of risks from time-dependent exposure patterns which are possibly
different from those used in the bioassay, but change with time. A
similar method was recently proposed by Thorslund.^^
Dose rates are often changed during the course of a bioassay,
although the changes often are not part of the original protocol. For
example, the higher experimental dose rates are sometimes lowered if
they appear to be causing acute toxicity. Likewise, dose rates are
sometimes increased during a study if it appears that the animals could
tolerate higher dosages. Some studies have variations in dose rates as
part of their original protocol. For example, some protocols specify
that dosing be terminated after about 1/2 of the animals' normal life-
spans have elapsed.In other studies, animals are given only a
single dose at the beginning of the experiment.^17^ Some studies (e.g.,
the EDQ1 study of 2-AAF)^®^ utilize different patterns of dose rates
for the express purpose of studying the effect of these patterns upon
carcinogenic response.
Any changes in the dose rate during the experiment, whether part of
the original protocol or not, complicate the analyses of the studies,
particularly if the analyses involve quantitative assessment of risk.
Some type of average dose rate typically is used as the dose measure for
purposes of risk assessment. However, this can give misleading results
because all doses are not equivalent; a dose applied early in life can
have a different effect than the same dose applied later, depending upon
the nature of the carcinogenic process. The procedures developed in
this paper offer a means of dealing with this situation within the
A-119
-------
-6-
framework of the multistage model. They utilize the exact dosing
pattern, rather than an average dose rate. They also can be used to
estimate risks under different exposure patterns than those utilized in
the experiment.
II. Basic Properties of the Armitage-Doll Model
If the dose rate is not constant, then equation (1) 1s modified by
replacing the average dose rate d, by the instantaneous dose rate D(t),
where D(t) is the dose rate to which an animal is exposed at age t. One
consequence of this assumption is that a carcinogen can have an effect
only when it is being administered. This is probably not a valid
assumption for chemicals which are stored in body fat and then released
slowly into the systemic circulation. For this reason and others, when
applying the model, it would be more appropriate to interpret D(t) as
internal dose at the site of action, rather than as externally adminis-
tered dose.
For an arbitrary dose pattern D(t), if several stages are dose-
related, the carcinogen response predicted by the multistage model has
an extremely complicated form. In the Appendix we have derived exact
expressions for the probability of a carcinogenic response by age t, for
the special cases in which either 1 or 2 stages out of k are affected by
the carcinogen and D(t) is a step-function. These special cases are
probably adequate for most purposes. Even if more than 2 stages are
dose-related,-it seems unlikely that there would ever be sufficient data
for distinguishing between 2 stages being affected and more than 2.
Likewise, any dose pattern D(t) can be approximated arbitrarily closely
by a step-function. Besides, dose patterns in most studies are step-
functions.
A-120
-------
-7-
If only the rth stage is dose-related the probability of a carcino-
genic response by age t is given by
P(t) « 1 - exp[-qQ(t -S)k - qrZrk(t - S)] (4)
(see Appendix) where qQ, qr, and S are parameters estimated by maximum
likelihood (these q's do not have precisely the same meaning as those in
(3J). The parameter^ represents a constant delay from cancer initiation
to expression. This delay is modeled as a constant for two reasons:
Firstly, the variability of this growth period is likely to be much less
than those of the initiation stages; secondly, to model realistically
this growth period as random would require the estimation of additional
parameters. The assumption that does not depend upon dose seems
appropriate at least for small doses.^ The first term in the exponent
(4) gives the incidence of spontaneous cancers, and the second term the
extra incidence due to the dose pattern D(t). The function Zr(c depends,
in a complicated manner, upon r, k, and the dose pattern D(t). A
similar expression (A23) holds whenever 2 stages are dose-related.
The formulae in the Appendix for constant exposure over a fraction
of a lifespan (equations A19-A24) enable us to consider the effect of
exposure periods of varying duration and beginning at different ages.
Tables 1-3 contain ratios, computed using these formulae, of the extra
risk of cancer by age 70 from a partial lifetime exposure at a given
level to the extra risk by age 70 from lifetime exposure at the same
level.* As an example of the interpretation of these tables, consider a
5-stage model and suppose that an individual is exposed from age 20 to
age 55 to 10 mg/kg/day of a carcinogen. If only the first stage is
affected by the carcinogen, according to Table 1, his additional
A-121
-------
-8-
lifetime risk 1s estimated to be 0.19 as large as that of an individual
exposed to 10 mg/kg/day for life.
These tables indicate that the extra risk from exposure depends
quantitatively upon the number of stages, which stage is affected by the
carcinogen, the duration of exposure, and the age at first exposure. If
the first stage is dose-related, early exposures are more dangerous than
later ones; according to Table 1, exposure for the first 35 years of
life gives almost the same risk as if the exposure continued throughout
life, provided there are at least 3 stages. For example, Table 1 shows
that with 4 stages the risk from 35 years exposure gives 94% of the risk
caused by lifetime exposure. On the other hand, if the penultimate
stage is affected, 20 years of exposure beginning at age 20 gives a
higher risk than the same exposure beginning at birth, and either a
higher or lower risk than the same exposure beginning at age 40, depend-
ing upon the number of stages.
III. Use of the Armitage-Doll Model in Carcinogenic Risk Assessment
We have developed computer programs that fit to bioassay data the
Armitage-Doll model with either one or two dose-related stages. The
program that utilizes 1 dose-related stage is called A-D0LL1 and the one
that utilizes 2 dose-related stages is called A-00LL2. The number of
stages k and which of these are dose-related are supplied by the user.
The programs also require, for each animal, the age at death (or the
1 These tables actually contain ratios of extra cumulative incidence H2
(second term in exponent of (4); see Appendix). However at low
doses extra cumulative incidence is a close approximation to extra risk.
A-12 2
-------
-9-
TABLE 1
RISK AT AGE 70 FROM 35 YEARS OF CONSTANT EXPOSURE EXPRESSED AS A
FRACTION OF THE RISK FROM 70 YEARS OF CONSTANT EXPOSURE
Number of Stages
Age at Beginning
of Exposure 3 4 5 6
Only First Stage Affected3
Same total
dose
0
1.75
1.88
1.94
1.97
20
0.71
0.52
0.37
0.26
35
0.25
0.13
0.06
0.03
Same dose
rate
0
0.88
0.94
0.97
0.98
20
0.35
0.26
0.19
0.13
35
0.13
0.06
0.03
0.02
Only Penultimate Stage Affected**
Same total dose
0
1.00
0.62
0.38
0.22
20
1.37
1.44
1.36
1.22
35
1.00
1.38
1.62
1.78
Same dose
rate
0
0.50
0.31
0.19
0.11
20
0.68
0.72
0.68
0.61
35
0.50
0.69
0.81
0.89
a Calculated using equation A19
b Calculated using equation A20
A-12 3
-------
-10-
TABLE 2
RISK AT AGE 70 FROM 20 YEARS OF CONSTANT EXPOSURE EXPRESSED AS A
FRACTION OF THE RISK FROM 70 YEARS OF CONSTANT EXPOSURE
Number of Stages
Age at Beginning
of Exposure 3 4 5 6
Only First Stage Affected3
Same total
dose
0
2.22
2.59
2.85
3.04
20
1.00
0.79
0.60
0.44
40
0.27
0.12
0.05
0.02
Same dose
rate
0
0.64
0.74
0.81
0.87
20
0.29
0.23
0.17
0.13
40
0.08
0.03
0.01
0.006
Only Penultimate Stage Affected^
Same total dose
0
0.69
0.26
0.09
0.03
20
1.43
1.24
0.92
0.64
40
1.18
1.66
1.96
2.10
Same dose
rate
0
0.20
0.07
0.03
0.01
20
0.41
0.35
0.26
0.18
40
0.34
0.47
0.57
0.60
a Calculated using equation A19
^ Calculated using equation A20
A-124
-------
-11-
TABLE 3
RISK AT AGE 70,FROM INSTANTANEOUS EXPOSURE EXPRESSED AS A FRACTION
OF THE RISK FROM SAME TOTAL DOSE PARCELLED OUT UNIFORMLY OVER 70 YEARS
Number of Stages
Age at Beginning
of Exposure
3
4
5
6
Only First Stage Affected3
0
3.0
4.0
5.0
6.0
20
1.5
1.5
1.3
1.1
40
0.55
0.31
0.17
0.09
60
0.06
0.01
0.002
0.0004
Only Penultimate Stage Affected^
0
0.0
0.0
0.0
0.0
20
1.22
0.70
0.33
0.14
40
1.47
1.68
1.60
1.37
60
0.74
1.26
1.80
2.31
a Calculated using equation A22
k Calculated using equation A23
A-12 5
-------
-12-
tlme at which a tumor was first observed, if different from the time of
death); a parameter related to the tumor status at time of death which
specifies what type of contribution the animal is to make to the likeli-
hood function; and the dose pattern to which the animal was subjected.
The dose pattern must be a step-function. Generally all animals in a
particular treatment group will be subject to the same dose pattern.
Using these data the program estimates the parameters qQ,qr,... and £ ,
which are described in the Appendix, by the method of maximum likeli-
hood. The user can provide to the programs an environmental exposure
pattern (which must also be a step-function), and compute estimates of
the extra cancer risk, at various ages, along with corresponding statis-
tical confidence limits, resulting from this exposure pattern. Esti-
mates of a dose scale parameter, and corresponding confidence limits, by
which the input dose pattern must be multiplied in %rder to achieve a
given level of extra risk can also be calculated.
The likelihood of the data may be represented as a product
where t.. is a time related to the ith animal and the product runs over
all of the animals in the experiment. The F for a particular animal can
be of 3 types:
corresponding to the event of interest not having occurred by time t^;
L =TTF(tJ
(5)
F(t.) = 1 - P(t,).
(6)
F(t.) = P(tA),
(7)
A-126
-------
-13-
corresponding to the event of interest having occurred prior to time t^;
and
F(tf) ¦ P(ti) = P'Uj). (8)
corresponding to the event occurring at time t^. Which of these 3
possibilities is used for a particular animal is determined by whether
the animal had a tumor, whether the tumor was incidental or fatal, and
the event whose probability is being estimated.
The Armitage-Doll probability P(t) may represent either the proba-
bility of death from a tumor by time t, or the probability of the
occurrence of a tumor by time t (i.e., the probability that a tumor has
been observed by time, or that one has occurred that would be detected
if a necropsy were carried out). If the probability of death is being
estimated, then animals having a fatal tumor contribute a term of the
type (8) to the likelihood, where t^ is the time of death, and animals
whose death Is not caused by a tumor (even though they may have a tumor
that is discovered incidentally at death) contribute a term of type (6).
This has the drawback that tumors discovered at terminal sacrifice are
not counted, even though these tumors may contribute the principal
evidence for carcinogenicity. If the probability of the occurrence of a
tumor is being estimated, then all animals with tumors contribute a term
of the type (7) to the likelihood, and tumor-free animals a term of type
(6). However, this analysis could be biased if the treatment has an
effect upon longevity. As an example to illustrate the possibility of
this bias, suppose there are no competing risks; i.e., all animals must
die from cancer. It can be easily shown that in this case the maximum
likelihood estimate of P{t) must be 1 for all times, even times prior to
the first cancer death. „
A-127
-------
-14-
Thus, both methods of analysis have drawbacks. The situation is
further complicated by the fact that in many studies no distinction 1s
made between fatal and incidental tumors. Our approach has been to make
analyses using different assumptions about whether tumors are fatal or
incidental. If these different analyses are comparable, one can have
more confidence in the results.
IV. Application
We Illustrate the method by applying it to the male rat data from
(19)
an NCI bioassay of ethylene dibromide (EDB), in which the dosage was
interrupted in the high dose group because of high mortality.
Fifty animals were placed into each of a high dose group and a low
dose group, and 20 animals were used as vehicle controls. Initially the
high and low dose groups received by gavage daily doses of 80 and 40
mg/kg, respectively. However, 19 of the rats in the high dose group had
died by the 15th week. Because of this high mortality, dosing to this
group was stopped during weeks 16-28. Dosing was resumed at week 29 at
the same level applied to the low dose group--40 mg/kg. Dosing was
halted in both groups during weeks 42 and 46. All surviving animals
were sacrificed by week 63. Figure 1 shows the Kaplan-Meier survival
curves for this experiment. Figure 2 shows the dose patterns and the
patterns of tumor occurrences in the 2 treatment groups. No tumors of
the forestomach were observed in control animals. The high Incidence of
tumors found during the last week of the study is due to tumors dis-
covered at final sacrifice.
The average doses in the two groups were quite similar—[(80)(15) +
(40)(19)3/49 = 40 mg/kg in the high dose group and [(40)(47)]/49 = 38
mg/kg in the low dose group. However, 45 out of 50 animals in the low
A-128
-------
Figure 1
Survival Curves for Male Rata Exposed to EDB
04
5 0.4
UMTMATtO CONTRIU.
MICH 0011
o a
OA
4S
X
7*
0
60
130
90
TIME ON TEST (WEEKS)
Source:
A-129
-------
-16-
Figure 2
Male Rats Exposed to E06 by Gavage
High Dose Animals
J.
Low Dose Animals
Weeks
A-130
-------
Figure 3a
Comparison of Estimates from Arm1tage-Do1l Model With Kaplan-Meier Estimates:
Assumed Six Stages. First Stage Dose Related
i
High dose:
Armitage Ootl
Kaplan Meier
£
Low Dose:
rmitoge Doll
Kaplan Meier
I
M9 o nlrr
-------
Figure 3b
Comparison of Estimates from Annitage-Doll Model with Kaplan-Meier Estimates:
Assumed Six Stages, First Stage Dose-Related,
Tumors Discovered During Week 49 Omitted
H igh
Armito ge Doll
Kaplan Meier^
Low dose:
-Arrrltage Doll
Kaplan Meier
10
50
15
20
40
30
-------
-19-
dose group had tumors (squamous-cel1 carcinoma of the forestomach), as
opposed to 33 out of 50 1n the high dose group. Clearly, it would be
inappropriate to base a risk assessment upon these average doses and
total numbers of tumors without taking into account competing risks, the
time pattern of tumor occurrences, and the unusual time-dependent dose
pattern.
This data set therefore is a candidate for analysis using the
Armitage-Doll model. Except for the animals that died during week 15 in
the high dose group almost all of the treated animals that died prior to
terminal sacrifice had a tumor. However, 4 animals in the low dose
group killed at terminal sacrifice were tumor-free. Therefore, it seems
likely that most of the deaths prior to terminal sacrifice in tumorous
animals were caused by the tumor. Consequently, we decided to focus
principally upon analyses in which probability of death from cancer was
estimated (i.e., analyses in which all animals with tumors contributed a
term of the form (8) to the likelihood). Upon fitting the models using
several different assumptions regarding the number of stages k, 1 £ k £
6, and which single stage was affected by the carcinogen, it appeared
that the best fit to the data (fit giving the largest likelihood) was
obtained assuming 6 stages with the first stage dose-related.
Figure 3a shows the Kaplan-Meier estimates of probability of death
from tumor, considering all deaths without tumors to be censored obser-
vations, compared to the corresponding probabilities estimated from the
best-fitting Armitage-Doll model (k = 6, first stage dose-related). The
Kaplan-Meier estimates show that, even though the total crop of tumors
is greater in the low dose group, tumors tended to occur earlier 1n the
high dose group, and, after correcting for competing mortality, the
A-134
-------
-20-
estimated probability of response is higher at the high dose than at the
low dose for each time point.
The assumption that the tumor caused the death is clearly inappro-
priate for tumors discovered at terminal sacrifice. To determine how
these tumors affected the fit, we refit the data ignoring tumors dis-
covered in week 49. As Figure 3b shows, this improved the correspon-
dence between the Kaplan-Meier curves and the curves derived from the
Armitage-Doll model. In fact the fit is excellent except for the
difficulty caused by the early cancer-related deaths in the high dose
group. There was a large number of tumor-free deaths which occurred at
the same time, and it appears likely that these early tumors were not
fatal. Although we did not apply a formal goodness-of-fit test, the
Armitage-Doll model appears to fit these data quite well.
It should be pointed out that it is appropriate to compare
f
Armitage- Doll and Kaplan-Meier of curves under the assumption that
tumors were fatal. However, if tumors had been assumed to be incidental
in fitting the Armitage-Doll model (i.e., if animals with tumors had
contributed a term of the form (7) to the likelihood) then comparison
with the Kaplan- Meier curves would have been inappropriate. Hoel and
Walburg^0^ graphs of the prevalence of incidental tumors can be used in
this circumstance.
Table 4 gives estimates of constant lifetime dose ratios corre-
-1 -2 -4
sponding to extra risks of 10 ,10 , and 10 based upon various
assumptions. Both maximum likelihood estimates and lower confidence
limits vary approximately linearly with dose. To illustrate the effect
of assuming different stages to be dose-related, we considered either
the first or third stage to be dose-related. As Table 4 indicates,
A-135
-------
-21-
TABLE 4
CONSTANT DOSE RATES CORRESPONDING
TO GIVEN LEVELS OF EXTRA RISK
LEVEL OF EXTRA RISK
MODEL
pr. of death from tumor by 3.3 0.31 (0.0031)
week 49; 6 stages; stage 1 (2.7) (0.26) (0.0026)
dose-related
pr. of death from tumor by 2.5 0.24 0.0024
week 49; 6 stages; stage 3 (2.1) (0.20) (0.0020)
dose-related
pr. of death from tumor by 4.7 0.45 0.0045
week 49; 6 stages; stage 1 (3.8) (0.36) (0.0036)
dose-related; tumors found
during last week of study
omitted
pr. of tumor by week 49; 6 0.91 0.086 0.00086
stages; stage 1 dose-related (0.70) (0.067) 0.00066
-5
pr. of death from tumor by 0.048 0.0046 4.6x10 c
week 105; 6 stages; stage 1 (0.037) (0.0035) (3.5x10 )
dose-related; tumors found
during last week of study
disregarded
a maximum likelihood estimates; 95S lower confidence limits in parentheses.
A-136
-------
-22-
doses estimated with these 2 approaches differed only by a factor of
about 1.3. This suggests that risks estimated from constant dose rates
are not extremely sensitive to the particular stage assumed to be
dose-related.
Because it is questionable whether tumors discovered at final
sacrifice should be used to estimate probability of cancer death, we
make estimates both including and excluding tumors found during week 49.
The differences in the estimates resulting from these 2 approaches was
about a factor of 1.4. However, this difference could be much larger in
studies in which nearly all of the tumors are discovered at terminal
sacrifice.
We also compared estimates of extra risk of tumor by week 49 with
comparable estimates of extra risk of death from tumor by week 49. The
former estimates are obtained by using terms of the type (7) rather than
(8) in the likelihood. These estimates are probably biased because of
the fact that most tumors likely were not discovered incidentally at
death, as assumed by the estimating method, but were the cause of death.
In spite of this difficulty and the fact that the 2 methods are estimat-
ing different quantities, the estimates do not differ greatly; estimates
of dose based upon time to tumor are smaller (as could be predicted) by
about a factor of 3.5 than comparable estimates based upon time to death
from tumor.
NCI rat bioassays typically last about 105 weeks, which is commen-
surate with the typical rat lifespan. However, because of the high
mortality in this study, the terminal sacrifice for male rats took place
much earlier. Some adjustment for such a short observation period is
needed when estimating lifetime cancer risks. For example, in their
A-13 7
-------
-23-
methodology for estimating water quality criteria, EPA (1980) adjusted
carcinogenic potencies upward by a factor (L/Lg) where L 1s the typical
lifespan of the species, and Lg is the duration of the experiment. If
the Armitage-Doll model is used for risk estimation, one can estimate
the risk at any age, including an age greater than the duration of the
experiment. In doing this, one is assuming that the observed response
pattern will continue past the period of observation. This calculation
is illustrated in the last row of Table 4 which contains estimates of
doses corresponding to risks after 105 weeks. These doses are almost
exactly a factor of 100 less than the comparable doses estimated for 49
weeks. This is reasonable for a 6 stage model because (105/49)^2: 100.
Interestingly, if the EPA^1^ correction had been used, the factor would
have been (105/49)3 = 9.8.
Finally, in Table 5 we examine the effect of a single week of
dosing which occurs at ei+her week 1, 14, or 30. It is estimated that a
dose given during the 14th week must be about twice as large as that
given during the first week to produce the same carcinogenic risk by
week 105 and a dose given during week 30 must be about 5 times as large.
V. Discussion
In this paper we have discussed some of the implications of the
Armitage-Doll multistage model of cancer to carcinogenic dose risk
assessment. We have emphasized particularly the effect of timing of
exposure upon cancer risk. The Armitage-Doll model predicts that a
particular exposure given early in life will have a greater carcinogenic
effect if an early stage is affected by the carcinogen than if a late
stage is so affected. We have also presented a new method, based upon
this model, for making quantitative risk estimates from animal bioassay
A-138
-------
-24-
TABLE 5
DOSES OF 1 WEEK DURATION THAT CORRESPONDS
TO GIVEN LEVELS OF EXTRA RISK3
BEGINNING LEVELS OF EXTRA RISK
OF EXPOSURE
10"1
Im
|o
1 1
1 ^
10~4
Week 1
0.91b
0.087
0.00087
(0.69)
(0.065)
(0.00065)
Week 14
1.8
0.17
0.0017
(1.4)
(0.13)
(0.0013)
Week 30
4.7
0.45
0.0045
(3.7)
(0.35)
(0.0035)
a Extra probability of death from cancer by 105 weeks; tumors found during
last week of study disregarded; 6 stages assumed with stage 1 dose-related.
k Maximum likelihood estimates; 952 lower confidence limits in parentheses.
A-139
-------
-25-
data. This method utilizes both the times at which cancers are observed
(or at which animals die if cancer-free) and the animals' specific
time-patterns of exposure. Using this data one can estimate carcino-
genic risk at any age and from under any exposure pattern.
Although we discussed different ways of dealing with the problem of
incidental tumors versus fatal tumors, none of these are entirely
satisfactory from a theoretical point of view. It appears that this
will usually not be a critical issue in light of the many other sources
of uncertainty in the risk estimates, since the results obtained from
the different approaches ordinarily will not differ appreciably. It
should also be kept in mind that similar, but more severe, difficulties
are present in estimates made from quantal data, although they are not
as apparent.
As is the case with any mathematical model, the multistage model is
an idealized representation of a carcinogenic process. It predicts that
cancer incidence will vary approximately as age to a power; this rela-
(2-3)
tionship is observed for many human carcinogens. ; No particular
phenomological interpretation is assumed for the cellular events leading
to cancer initiation, although it is reasonable to assume that some of
them are mutational in nature. It has recently been demonstrated that 2
point mutations can be involved in human carcinogenesis.{21-22)
suggests a process with at least 2 stages. These point mutations can
apparently occur in any order, whereas in the particular multistage
model implemented in this paper the stages must be sequential. However,
it would not be difficult to consider a version of the model in which
the events did not have to occur in a particular order. If the events
A-140
-------
-26-
separating stages are mutations, it would seem reasonable that there
would be a linear relationship between internal dose—that is, the dose
to the target tissue—and the mutation rates, which is in agreement with
the assumption in the multistage model as illustrated in equation (1).
For example, if the mutation is an incorrect base substitution the
mutation rate should be proportional—at least at low doses—to the
number of abberant base molecules available, which in turn should be
proportional to the internal dose.
The Armitage-Dol1 model is more appropriately applied to the dose
to the target tissues rather than to the external dose to which the
animals are exposed; pharmacokinetic mechanisms relating external dose
to tissue dose are not incorporated into the model. Frequently,
carcinogenesis dose-response data exhibit a high degree of non-
linearity. To achieve this non-linearity within the Armitage-Doll model
requires several dose-related stages. It seems likely, however, that
most of this non-linearity is due to non-linearity in pharmacokinetic
(23)
mechanisms* ' rather than in mechanisms of carcinogenesis per se. For
example, once non-linearities in uptake are accounted for 1n an animal
bioassay for vinyl chloride the data became consistent with a one-hit
/25)
model. " ' It seems entirely possible that, once the non-linearity
in pharmacokinetic mechanisms are accounted for, Armitage-Doll models
with a single dose-related stage may be adequate models for most carci-
nogenesis dose-response data.
There are both observed phenomena associated with particular
carcinogens or cancer types, as well as theories of cancer, which are
not incorporated into the multistage model. For example, the process of
DNA repair is not included in the model, nor is the effect of the immune
A-141
-------
-27-
system. It is possible that the principal effect of DNA repair upon the
dose response would be to reduce the rate at which mutations occur. If
so, then inclusion of DNA repair in the model would lead to larger
estimates of or and ^ which would be offset by the effect of DNA
repair, but would not substantially affect the overall predictions of
the model. If DNA repair or other more detailed mechanistic assumptions
were introduced into the model one would have to deal with the problem
of estimating additional parameters from data which are possibly already
overburdened. A currently popular view of carcinogenesis is that it is
an initiation-promotion process. This view could be considered within
the multistage approach, with initiation relating to early stages and
(14>
promotion to later stages. ' However, the promotion stage—at least
as observed in animal skin painting studies—can exhibit regression,
with tumors becoming smaller or disappearing altogether when the treat-
ment with a tumor promoter is stopped. This behavior is not modeled in
the current model; stages, once attained, are irreversible and lesions
are not allowed to regress on their way to becoming cancerous.
Thus the simple multistage formulation in this paper cannot be
expected to represent many detailed features of complicated and little-
understood carcinogenic processes. However, it is successful in de-
scribing many epidemiological and experimental observations. It seems
to us to provide a useful framework for investigating the quantitative
aspects of time-dependent dosing.
Acknowledgement: The authors would like to express appreciation to
Dr. Todd Thorslund and others in the EPA Carcinogen Assessment Group
(CA6) for helpful discussions, and for making their related work avail-
able to the authors. Although the research described in this article
A-142
-------
-28-
has been funded by the United States Environmental Protection Agency
through contract 68-01-3111, it has not been subject to the Agency's
peer and administrative review and therefore does not necessarily
reflect the views of the Agency, and no official endorsement should be
inferred.
A-14 3
-------
-29-
References
1. P. Armitage and R. Doll, Stochastic models for carcinogenesis. In:
Proceedings of the Fourth Berkeley Symposium on Mathematical
Statistics and Probability. Vol. 4, pp. 19-38. (University of
California Press .'Berkeley, California, 1961).
2. R. Doll, The age distribution of cancer: implications for models
of carcinogenesis, Journal of the Royal Statistical Society, Series
A, 134:133-166 (1971).
3. R. Peto, Carcinogenic effects of chronic exposure to very low
levels of toxic substances, Environmental Health Perspectives
22:155-161 (1978).
4. K. S. Crump, D. G. Hoel, C. H. Langley, and R. Peto, Fundamental
carcinogenic processes and their implications to low dose risk
assessment, Cancer Research 36:2973-2979 (1976).
5. H. A. Guess and K. S. Crump, Low-dose extrapolation of data from
animal carcinogenesis experiments—analysis of a new statistical
technique, Mathematical Biosciences 32, 15-36 (1976).
6. K. S. Crump, H. A. Guess and K. L. Deal, Confidence intervals and
tests of hypotheses inferred from animal carcinogenicity data,
Biometrics 33, No. 2, 437-451 (1977).
A-144
-------
-30-
7. H. A. Guess and K. S. Crump, Maximum likelihood estimation of
dose-response functions subject to absolutely monotonic con-
straints, Annals of Statistics 6, No. 1, 101-111 (1978).
8. K. S. Crump, An Improved procedure for low-dose carcinogenic risk
assessment from animal data, Journal of Environmental Pathology and
Toxicology, 5(2):675-684 (1981).
9. K. S. Crump, Statistical aspects of linear extrapolation. In:
(ed. C. R. Richmond, P. J. Walsh, and D. Copenhaver). Proceedings
of the Third Life Sciences Symposium, Health Risk Analysis, pp.
381-392., (Gatlinburg, Tennessee, 1981).
10. K. S. Crump and R. B. Howe,. Review of Methods for Calculating
Confidence Limits in Low Dose Extrapolation. In: (Krewskl, D.
ed.) Toxicological Risk Assessment, (CRC Press, Inc.rCanada, 1983)
(in preparation).
11. EPA, Water Quality Criteria Documents; Availability, Federal
Register 45, No. 231 (Friday, November 28), 79317-79379 (1980).
12. M. S. Cohn, Revised Carcinogenic Risk Assessment for Urea Formalde-
hyde Foam Insulation: Estimates of Cancer Risk Due to Inhalation of
Formaldehyde Released by UFFI (1981).
A-14 5
-------
-31-
13. A. S. Whittemore and J. B. Keller, Quantitative theories of carci-
nogenesis, Society for Industrial and Applied Mathematics Review
20:1-30 (1978).
14. N. E. Day and C. C. Brown, Multistage models and primary prevention
of cancer, J. Nat. Cancer Inst. 64:977-989 (1980).
15. T. W. Thorslund, Estimation of the Effects of Exposure to a
Carcinogen that Fluctuate Over Time on the Lifetime Risk of Cancer
Death. Presented at the Pacific Division, AAAS, Annual Meeting
(1983).
16. C. Maltoni, G. Lefemine, A. Ciliberti, G. Cotti and D. Carretti,
Carcinogenicity bioassays of vinyl chloride monomer: a model of
risk assessment on an experimental basis, Environmental Health
Prospectives 41:3-29 (1981).
17. M. F. Stanton, M. layard, A. Tegeris, E. Miller, M. May, and E.
Kent, Carcinogenicity of fibrous glass: Pleural response in the rat
in relation to fiber dimension. J. National Cancer Institute
58:587-597 (1977).
18. T. Cairns, The Edpj Study: Introduction, objectives and experi-
mental design, Journal of Environmental Pathology and Toxicology
3(3):1-7 (1980).
A-146
-------
-32-
19. NCI, Bioassay of 1,2.Dibromoethane for Possible Carcinogenesis.
Technical Report Series No. B6, DHEW Publication No. (NIH) 78-1336
(1978).
20. D. G. Hoel and H. E. Walburg, Statistical analysis of survival
experiments, J. National Cancer Institute 65:361-372 {1972).
21. P. McGrath, D. J. Capon, D. H. Smith, E. Y. Chen, P. H. Seeburg, D.
V. Goeddel and A. D. Levinson, Structure and organization of the
human Ki-ras proto-oncogen and a related processed pseudogene,
Nature 304, 501-506 (1983).
22. D. J. Capon, P. H. Seeburg, J. P. McGrath, J. S. Hayflick, U.
Edman, A. D. Levinson and D. V. Goeddel, Activation of Ki-ras3 gene
in human colon and lung carcinomas by two different point
mutations, Nature 304, 507-513 (1983).
23. D. G. Hoel, N. L. Kaplan and M. W. Anderson, Implication of non-
linear kinetics on risk estimation in carcinogenesis, Science
219:1032-1037 (1983).
24. P. J. Gehring, P. G. Watanabe and C. N. Park, Resolution of dose-
response toxicity data for chemicals requiring metabolic
activation: example-vinyl chloride, Toxicology and Applied
Pharmacology 44, 581-591 (1978).
A-14 7
-------
-33-
25. SRS, An Investigation of the Use of Pharmacokinetic Data in
Carcinogenic Risk Assessment, Prepared for the II. S. Environmental
Protection Agency, Contract No. 68-01-5975, Task A, Subtask No. 6,
Issue No. 3 (1981).
26. 0. R. Cox and D. V. Lindley, Theoretical Statistics (Chapman and
Hall, London, 1974).
27. R. B. Howe and K. S. Crump, GLOBAL 82: A Computer Program to
Extrapolate Quantal Animal Toxicity Data to Low Doses. Prepared
for the Office of Carcinogen Standards, OSHA, U.S. Department of
Labor, Contract 41USC252C3 (1982).
A-14 8
-------
-3U-
APPENDIX
Al. Derivation of special cases of the Armitage-Doll model
In the Armitage-Doll model of cancer, a tumor results when k events
have occurred in a single cell. Within each cell these events
occur in order. Once the first i - 1 events w^...,#^ have occurred,
the ith event occurs at the rate ^(t), where t represents age (not
the time since the occurrence of the i-lst event). The dose rate D(t)
at age t of a carcinogen is introduced into the model by assuming that
^(t) = +piD(t). (Al)
The parameter ^ represents the rate at which the ith event occurs in
the absence of a carcinogenic insult and is assumed to be age-independent.
The parameter^ represents the carcinogenic potency of the carcinogen
upon the ith stage, and is likewise age-independent.
Under these assumptions it can be shown that the probability
density of the age at which w2 occurs in a particular cell is given by
u2 u1 u2~ul
f2(u2) = S X^UjJexpf-i A1(v)dvlX2(u2)exp[-5 + v)dv]dUj
(A2)
s ^"^(u^1^2(^2)R2(,u2)du^,
The last equality defines R2> Note that R2(u^,u2) —* 1 uniformly for
0 5 u1 f u2 f t as l(ul) * ^21u2^ both ^ 0 uniformly in this inter-
val. By induction it can be shown that the probability density of the
age at which the ith event w^ occurs in a single cell can be written as
A-149
-------
-35-
f^(u.j) j ••• 5 A1("1)-.Ai(ui)Ri(u1,...,ui)du1,
i^VVV
• • •»
u i)du^ ,.«. ,du^
for i = 2,... ,k, where
R.j(uj,...,u^) —^ 1 uniformly for 0 *> Uj ? ...iu^ £ t as
Ai(Ui)..Xj(ui) -^> 0 uniformly. The probability Pc(t) of a
1 • • • I
(A3)
cancer in a particular cell line by age t is the integral of the probabi-
lity density
In a tissue comprised of n cell lines acting independently, the
probability of a cancer by age t is
Since n is typically a very large number, in order that P(t) is a
moderate number the probability Pc(t) that cancer results in a partic-
ular cell line must be small. One way to express this mathematically
(there are many other possibilities that would lead to essentially the
same conclusion) is to suppose thato^. and f^ depend upon n in such a
1/k l>k
way that «^^n ' —* a^ andji^n —^ b^ as n —» It then follows
that^.(u^) —^ 0 uniformly in any finite interval and thus the limiting
process and integration can be interchanged. Using (A3), (A4) and (A5)
we get Pn(t) —» P(t) as n —* ®<* , where
(A4)
Pn(t) - 1 - [1 - Pc(t)]n
° 1 - exp£nLn[l - Pc(t)]j.
(A5)
P(t) = 1 - exp[- H(t)]
(A6)
A-150
-------
-36-
and the cumulative hazard H(t) is given by
C "" (! ^ + biDtu1)3 — Cafe + bkD(uk)3du1...duJ(. (A7)
This expression will be our starting point for deriving expressions for
P(t) corresponding to various special cases.
Al.l. Case in which a single stage is dose-related: Suppose that the
rth stage only is dose-related. This means that br> 0 for a particu-
lar r, l£ r 6 k and that bj 8 fl for i f r. Then H(tJ can be written
as the sum of 2 integrals H^(t) and Hg(t) where
1 «k r2
Mt) ¦ (if a.) f J ... S du, ...dut = (TI a.)tk/k! (A8)
i 1 o 0 0 1 K 1
represents the background cumulative incidence and
H2(t) = C(iraj)(br/ar)3Zr|c(t)/k! (A9)
represents the extra cumulative incidence due to exposure, where
:! ^ ... 1} D(ur)dUj..
2 (t) = k! J J ... j D(uJdUT-.du. (A10)
0 0 0
To evaluate Zr^(t)» we integrate first with respect to u1...ur_1,
and then--after reversing the order of integration—with respect to
The result is
A-151
-------
-37-
Zrk(t) iD{u_)(t - uJ^VV-
r* (k-r)l(r-l)! 0 r r r r
L.r
Applying a binomial expansion to (t - up) yields
k-r t
Zrk(t) = 2 (-l)Jk! tk"r'J f D(ur)urj+Mdur (A12)
j! (r-1)!(k-r-j)1
In order to proceed further we must make more detailed assumptions
regarding the dose rate function D(t). It will be sufficient to con-
sider D(t) to be a step-function because any reasonably well-behaved
function can be approximated by such a function. Additionally, dose
patterns in animal experiments typically are step-functions. Therefore,
we suppose that D is the step-function
D(x) « hi, si_1 5 x < si (A13)
for 1 2 0, where sn = 0. For fixed t, let m satisfy sm , < t < sm. Then
u J m-i m
+ -«£!))
¦ - h1H)sfr * VJ+rJ (A14)
y— ~ h^J
where the last equality defines the function G.
Expressions (A6), (A7), (A8), (A10), (A12) and (A14) define a
closed form expression for P(t). In the special case m = 1 (constant
dose of hj) this solution reduces to the Weibull form
A-152
-------
-38-
P(t) - 1 - expC>h1t!Ta1)U + b^a^/k!] (A15)
just as it should.
In the computer implementation of the model a delay £ is included
which represents the period from the initiation of the tumor to its
clinical expression. Modifying the definition of P to incorporate this
delay, the form of the model implemented by the computer is
P(t) = 1 - exp[-qQ(t - £)k - qrZpk(t-S)] (A16)
where
q0 = (ffaf)/k! (A17)
and
qr = (*Ta1 )(br/ar)/k!. (A18)
The computer program obtains maximum likelihood estimates of the 3
parameters of qg, qr» and $, all of which are constrained to be
non-negative. The parameters k and r are fixed in advance and not
estimated.
The special case in which exposure is at a constant rate d in the
interval [Sj, Sg] and is zero elsewhere is of particular interest.
It can be shown, either by using (AI2) and (A13} or by integrating (All)
directly, that
A-153
-------
-39-
H2(t) = db1C7Ta1) ^
k! a.
(t - Sl)'
t « s,
sl' t< s2
(A19)
(t - s2)k - (t - s2)k s2 6- t
if the first stage 1s affected ( r « 1),
H2(t)
= dbk.i^.)
tk - sk~l[kt - (k - l)Sl]
k > a
k-l
sJS'^kt - (k-l)s2] - Sj"A[kt - (k-l)s13
k-l,
t ^ s.
Sj £ t <¦ S2
S2 S t.
(A20)
if the penultimate stage is affected ( r ° k - 1), and
H2(t) = db^taj)
k!a.
t < Si
Sj i t ^ s2 (A21)
s2i t
if the last stage is affected (r = k).
If exposure is instantaneous (or nearly so) at time Sj to a total
dose D, the subsequent risk can be derived by taking limiting cases of
the above results. The expressions obtained are:
Hz(t) « Db1(ITai) (t - Sj)
(k-Dlaj
k-l
t >s
1
(A2
for the case for which only the first transition is affected (r = 1);
A-154
-------
-140-
H2(t) • Dbk_1(JTa1) sj*2(t - Sj) t * Sj (A23)
Sj (A24)
(W)!ak
for the case for which only the last transition is affected (r 8 k).
A1.2 Case in which two stages are dose-related: Suppose stages r and
(r< i) are dose-related; that is, bp ? 0, > 0, and b. = 0
otherwise. The hazard function (A7) can now be written
.?2
H(t) ) ) ... ) [1 t (br/ar)D{ur) + (tj|/aft)D{us) (A25)
+ brb
-------
-1*1-
t u^ U2
Zr,t(t) = klj J ...\ 0(u<)0{ur)du1...dulr (A28)
'rtr ' 0 0 "0 "*"«'^"r'""l'—"k
- «f
0 0 0
D(Vznt-i'u.£)dy "duit'
Using the expressions developed previously for and reversing the
order of integration, we reduce this expression to the single integral
Zrck^ = $ % 7)) ki [G(j+r,x-l)u, r 0
r,lc 0 j=0 J! (r-1 J i (p -1-r-j j! (j+r) *
(A29)
+ Vj"1] D(uf)(t - dug
(k-jr)!
where x is the function of defined by sv ,£ n < sv. The term
X X-l g X
G(j+r, x-1) is thus a step-function in that has jumps at the same
points as D(u^). The integral (A29) is evaluated by expanding
L A
(t - u ) ' in a binomial series. The result is
t
k-Si-l-r
i=0 j=0 i!j!li-l-r-j]!(k-A-i)i(r+j)
^ ("[^(r+j.n-l) - hn+1G(r+j,n)]sJ+1"r"'i
n=l j
I X +i-r-j
(A30)
i +i
+ hmG(r+j, m-l)ti+1"r"j + hV+1
J +i-r-j
Jl +1
A-156
-------
-U2-
where m satisfies sm_^ £ t C Sm. The implementation of this equation
by the computer is not as difficult as it looks, as a number of the sums
can be calculated a single time and recalled when needed.
The delays is included here also and consequently the computer
program estimates the 5 parameters qQ, qpl qtf, qpi, and S, each of
which is constrained (only) to be non-negative. Note that this is a
more generalized form than that predicted by the Armitage-Doll theory
because it does not include the constraint q^qg = qr<^ which is implied
by equations (A17), (A18), and (A27).
A2. Description of Computer Programs
The parameters qg, qr,..., and S are estimated by maximizing
the likelihood of the data. Denote the dependence of the probability
P(t) upon the dose pattern d(t) by writing P(t) = P(t;D). Parameters of
interest 1n risk assessment include the extra risk by a given age t from
a given dose pattern D defined as
where IT is a specified amount of extra risk and D is a specified dose
pattern. Thus, f is the amount by which a given dose pattern D must be
multiplied to produce an extra risk of II.
Confidence limits for extra risk and f are computed using the
asymptotic distribution of the likelihood ratio statistics.(26-27)
R(t;D) = P(t;D) - P(t;0)
1- P(t;0)
(A31)
and the dose scale parameter f defined as the solution to
(A32)
A-15 7
-------
-1»3-
For example, the 95% lower confidence limit for f is defined as the
minimum f which satisfies (A29) and
" L> * I1"645'2
where L Is the "log-11 kelihood of the data and Lmax is the maximum
value of the log-likelihood.
A-258
------- |