Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.

Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 202

F
Development Toxicology
Many hazardous environmental chemicals seem to have toxic effects in the
offspring of the persons exposed, including bird defects and spontaneous
abortion. Therefore, in assessing the risks associated with chemical mixtures,
one should investigate their possible teratologic effects. As a step in assessing
these risks, animal bioassays are performed at increasing doses of the chemi-
cals to be studied. The end points of interest in the bioassays are fetal death,
failure to grow, structural and functional abnormalities, and behavioral defi-
ciencies. Other elements of the reproductive cycle are not considered.
A typical animal bioassay is carried out by exposing a female before or
during pregnancy or a male before mating, observing their offspring (if any),
and studying the end points in relation to the doses administered. The toxic
responses are recorded in the offspring, but the experimental units are the
females. Mantel (1969) recognized that an inherent characteristic of this type
of data is the so-called litter effect, which is the tendency for littermates to
respond more similarly than animals from different litters. As a result, the
quantal dose-response information obtained from teratologic experiments is
different from the usual dose-response data, based on effects (tumors, etc.)
observed in directly exposed animals. The methods of analyzing such data with
dose-response models must take the difference into account. We review these
methods briefly here.
STATISTICAL METHODS FOR
DEVELOPMENTAL`-TOXICOLOGY DATA
In the typical developmental-toxicology experiment with animals, the data
collected are expressed as in Table F-1.
202

OCR for page 202

APPENDIX F
TABLE F-1 Characteristics of Data from Developmental-Toxicology
Experiments with Animals
Test Groups
203
Control Groups
=0
Low
High
Dose
No. females
No. offspring
No. offspring with defect xOI, . . ., xO,~O
dl
no
Sol ~ . . · ~ SOno
Al
Sl I ~ . · · ~ Sln
Xl I, . . ·, Xln
dm
Em
Sml' · ~ Smam
Xml, . . · ~ Smn/
As Table F-1 suggests, the experiment is usually carried out at m increasing
doses do < . . . < dm in the control group, at which there is no exposure,
do = 0. The experimenter randomizes ni females to the ith group and exposes
them to dose di. For each female in dose group di, the experimenter records the
litter size sir, with j = 1, 2, . . . , ni, and xij the number of offspring with a
specified toxic defect. Let pij be the probability that an offspring in the it litter
will be defective. In analyzing such quantal response data statistically, the
probability pij is usually assumed to vary from female to female. Within a litter,
the number of defects, xij, is assumed to have a binomial distribution with
defect rate pin for the sin offspring. Such modeling allows for estimating within-
litter and between-litter variation.
Haseman and Kupper (1979) and Van Ryzin (1985a) have discussed a num-
ber of statistical procedures for such a model. In particular, one can estimate
the mean defect rate at each dose (Van Ryzin, 1975), test whether a particular
dose group differs from the controls in defect rate (Hoer, 1974), and test
whether the mean defect rate increases with dose by using Jonckheere's non-
parametric trend test (see Lehman, 1975, p. 233~. A parametric approach was
proposed by Williams (1975), who fitted the beta-binomial model to terato-
logic data. In still another approach, Kupper and Haseman (1978) presented a
correlated binomial model, which, in contrast with the beta-binomial model,
allows for negative as well as positive intralitter correlation. In both previous
cases, the authors estimate the relevant parameters with maximum-likelihood
methods. A likelihood ratio test is used to compare results in the treatment and
control groups. (For further discussion of the procedures, see Kupper and
Haseman, 1978; Van Ryzin, 1985a.)
The statistical procedures described make it possible to test for the presence
or absence of a dose-response relationship in developmental-toxicology bio-
assays for various developmental end points or for fetal toxicity. It is also
possible to assess the effects of dose on conceptus resorption, which is thought
to be the rough equivalent in animal studies of spontaneous abortion in hu-
mans. That can be done by standard analysis methods in which sit is defined as
the number of implants of them female in dose group di end xij is the number of
resorbed fetuses.

OCR for page 202

204
APPENDIX F
As is the case for directly observable quantal response data, such as carcino-
genicity data, the establishment of a dose-response relationship raises the ques-
tion of how to estimate the risks associated with low environmental exposure to
a substance, given the responses observed at (usually) higher experimental
doses. That involves low-dose extrapolation through the use of some dose-
response model. The models for developmental-toxicology data proposed in
the literature are discussed in the next section, which also includes a new test
for detecting a response trend with dose for teratology data.
DOSE-RESPONSE MODELS
FOR DEVELOPMENTAL EFFECTS
The literature on development of dose-response models for developmental
effects or fetal death is rather sparse. A broad set of models must be considered,
because no process to explain the effects is generally accepted, and many end
points are involved. It is commonly held that developmental effects and fetal
toxicity have a threshold dose-response relationship and that safety factors can
be used to protect public health. However, the applicability of threshold
models for all developmental end points is open to debate. For example, Jusko
(1972, 1973) examined two classes of developmental toxicants those that
appear to have a threshold and those that do not. In particular, Jusko's analyses
(1972, 1973) would indicate that thalidomide and cyclophosphamide are non-
threshold developmental toxicants.
At least two attempts to model dose-response relationships for developmen-
tal effects have been reported. Jusko (1972) has argued for a drug receptor
model for fetal exposure based on pharmacokinetics. Assuming a "hit theory,"
he modeled the fraction of intact or normal fetuses as an exponentially decreas-
ing function of dose for some toxicants. In the notation used here, sin is the
number of implants at dose di in female j, and xij is the number of resorbed,
dead, or defective fetuses from this female. Thus, the response rate (live
births), (sin—xij~lsii, of normal fetuses would have a mean or expected value of
the form
mean [(sin—xii~lsii] = exp ~—Kdi),
for K > 0, j = 1, . . ., ni, and i = 0, 1, 2, . . ., m. Equation F-1 assumes that
there is no background response rate, but if there were one, it could be included
by substituting exp ~—Kdj + do) in the right side of the equation. For the no-
background-rate case, Equation F-1 can be rewritten as
mean (1—Pij) = exp (-Kdi),
(F-1)
(F-2)
where Pin is the proportion of dead, resorbed, or defective fetuses in the off-
spring of the JO female exposed to dose di.

OCR for page 202

APPENDIX F
205
Jusko (1972) argued further that other developmental toxicants that must
affect r receptor sites (r > 1) before having a toxic effect in offspring would
follow Equation F-3:
mean Isis—xij~lsij] = 1 - [1—exp ~—Kdi)]r, (F-3)
where r > 1 and no background effect is assumed. Equations F-1 and F-3 may
be rewritten as one equation in terms of the mean defect rate at dose di:
Oldie = mean (xij/sij) = [1—exp ~—Kdi)]r, (F4)
for r—1. A model like Equation F4 could be used for low-dose extrapolation
by extending methods of low-dose extrapolation now in existence for directly
observable quantal dose-response models (Krewski and Van Ryzin, 19811.
However, such methods have not yet been fully reported. When case r = 1,
Equation F4 implies that the dose-response is linear at low doses (low re-
sponses); that is, for d near zero, Equation F4 becomes
~(d)=1—expf—Kd)~Kd.
(F-5)
Equation F-5 implies that at times linear extrapolation at low doses is appropri-
ate. Assuming that the above model is valid, one could combine the low-dose
effects of simple mixtures in a manner similar to that suggested for carcinogen-
esis models (see Appendix E).
A second attempt at low-dose extrapolation with dose-response models for
developmental toxicology data has been proposed by Rai and Van Ryzin
(19851. They modeled the variation of the defect rate or proportion Pij by
taking
Pij = \(di) exp ~—Two + . · · + hkwki)], (F-6)
wh~.re. low.. ~ ~ R.u'.. ~ n ~ - Up \tA.N rlPn~t~c, ~ hackling ~ - fort Ate Or
l lJ ~ · · · r~K~K} — an- · - __ ~~~ _--_ ~ v_--_ ~~_~ $~— ~$
proportion at dose di, and the exponential term represents the modification of
that baseline rate due to the particular covariates or mitigating factors Wvj'
v = 1, . . ., k, for thejth female. For example, one could include such charac-
teristics of the dam as maternal weight loss and litter size as covariates in this
model. However, the covariates, Wvj, need to be observable. The baseline re-
sponse \(d) in Equation F-6 also needs to be specified. Rai and Van Ryzin
(1985) studied a particular case of Equation F-6 by taking
\(d) = 1—expE—(a + ~d)]
(F-7)
for or > 0 and ~ > 0 and including a single covariate, the litter size (so). If
Equation F-7 holds, the female baseline defect-response rate follows a one-hit
model. In this case, low-dose extrapolation based on X(d) would be linear.
Extensions of the work of Rai and Van Ryzin could be developed for low-dose
extrapolation based on any specified \(d ~ (Van Ryzin, 1985a). If Equation F-7

OCR for page 202

206
APPENDIX F
is appropriate and low-dose extrapolation for each chemical is based on X(d),
corresponding low-dose extrapolation for mixtures would be dose-additive, as
is the case for the multistage model in carcinogenic studies (see Appendix E).
A related problem was addressed by Munera (1986), who proposed a trend
test for teratologic data. In this approach, the random response probability, Pij,
is allowed to vary within the dose group di according to an unknown distribu-
tion, Gi. However, the mean ofthe distribution is assumed to follow a specified
regression function that might depend on dose. For example, if Equation F-2
holds, one could take
mean (Pij) = 1—e(-K~i)
The statistical trend test derived from this model is based on empirical
Bayesian statistical procedures applied to the unknown distribution Gi and
tests whether the regression function increases with dose.
LIMITATIONS AND ADVANTAGES OF PROCEDURES
THAT USE DEVELOPMENTAL END POINTS
The statistical procedures for analyzing developmental-toxicity studies dis-
cussed previously try to account for the greater than binomial variability found
in the data collected from such experiments. However, no model seems to have
a clear advantage over the others (see Haseman and Kupper, 19791. Little
effort has been made to compare the advantages of the various approaches
systematically.
The trend test proposed by Munera (1986) is also new. An alternative way of
estimating the regression parameters might be more efficient than their empiri-
cal Bayesian estimates. This method merits further investigation.
Similarly, the main problem in using the dose-response models mentioned in
their article is that they are not yet fully developed and have not yet received
general scientific acceptance. That is, the dose-response models proposed for
developmental data are not yet clearly justified by the biologic evidence. The
models discussed above appear to fit some data well, but it is possible that
no single dose-response model for developmental toxicity will be generally
acceptable.
The results to date do suggest that several developmental toxicants fall into
the class of materials exhibiting low-dose linearity. If the parallel to carcino-
genesis holds where nonlinearities in the observed dose range due to nonlin-
ear kinetics are consistent with low-dose linearity (Hoer et al., 1983; Rai and
Van Ryzin, 1984; Van Ryzin, 1985b) then low-dose linear extrapolation
would be appropriate in assessing the risk of developmental effects.
Data on developmental toxicity of mixtures and their components can be
gathered more quickly than data on carcinogens, inasmuch as animal reproduc-
tive-developmental bioassays can be carried out relatively rapidly (about 2

OCR for page 202

APPENDIX F
207
months), as opposed to carcinogenicity assays (approximately 2 years). Thus,
the possibility of economical testing of a variety of mixtures for developmental
toxicity is greater than the possibility of economical testing for carcinogenicity.
SUMMARY
Developmental-toxicity end points present an opportunity for assessing He
joint actions of chemicals. In the usual laboratory expenment, each exposed
female provides several offspnng, so both litter size and incidence of abnor-
malities within litters provide an index of toxicity. In addition, the incidence of
resorptions can tee assessed. Dose-response iCunctions can tee extracted from the
venous end points, but the crucial problem is low-dose extrapolation. Models
that have been devised for that purpose suggest that low-dose extrapolation for
mixtures would likely be dose-additive, but there is no general acceptance of
these models.
REFERENCES
Haseman, J. K., and L. L. Kupper. 1979. Analysis of dichotomous response data from certain toxico-
logical experiments. Biometrics 35:281-293.
Hoel, D. G. 1974. Some statistical aspects for experiments for determining the teratogenic effects of
chemicals, pp. 375-381. In J. W. Pratt, Ed. Statistical and Mathematical Aspects of Pollution Prob-
lems. Marcel Dekker, New York.
Hoel, D. G., N. L. Kaplan, and M. W. Anderson. 1983. Implication of nonlinear kinetics on risk
estimation in carcinogenesis. Science 219: 1032-1037.
Jusko, W. J. 1972. Pharmacodynamic principles in chemical teratology: Dose-effect relationships. J.
Pharmacol. Exp. Ther. 183:469-480.
Jusko, W. J. 1973. Pharmacokinetic principles in chemical teratology, pp.9- 19. In W. A. M. Duncan,
Ed. Toxicology: A Review and Prospect. Proceedings, European Society for the Study of Drug
Toxicity. Vol. 14. American Elsevier, New York.
Krewski, D., and J. Van Ryzin. 1981. Dose response models for quantal response toxicity data,
pp. 201-231. In M. Csorgo, D. A. Dawson, J. N. K. Rao, and A. K. Md. E. Saleh, Eds. Statistics
and Related Topics. North-Holland, New York.
Kupper, L. L., and J. K. Haseman. 1978. The use of a correlated binomial model for the analysis of
certain toxicological experiments. Biometrics 34:69-76.
Lehman, E. L. 1975. Nonparametrics: Statistical Methods Based on Ranks. Holden Day, San
Francisco. (457 pp.)
Mantel, N. 1969. Some statistical viewpoints in the study of carcinogenesis. Prog. Exp. Tumor Res.
11:431~43.
Munera, C. 1986. An empirical Bayes approach to risk assessment in certain reproduction studies.
Ph.D. dissertation, Columbia University.
Rai, K., and J. Van Ryzin. 1984. A dose-response mode incorporating Michaelis-Menten kinetics,
pp. 59-64. In American Statistical Association, Biopharmaceutical Section. Proceedings. American
Statistical Association, Washington, D.C.
Rai, K., and J. Van Ryzin. 1985. A dose-response model for teratological experiments involving
quantal responses. Biometrics 41: 1-9.
l

OCR for page 202

208
APPENDIX F
Van Ryzin, J. 1975. Estimating the mean of a random binomial parameter with trial size random.
Sankhya Ser. B. 37:10-27.
Van Ryzin, J. 1985a. Consequences of nonlinear kinetic dose-response models on carcinogenic risk
assessment, pp. 119-132. In D. G. Hoel, R. A. Merrill, and F. P. Perera, Eds. Risk Quantitation and
Regulatory Policy. Banbu~ Report 19. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y.
Van Ryzin, J. 1985b. Risk assessment for fetal toxicity. Toxicol. Ind. Health 1:299-310.
Williams, D. A. 1975. The analysis of binary responses from toxicological experiments involving
reproduction and teratogenicity. Biometric 31:949-952.