f iff Systems. Jt\
Submitted to:
Office of Pesticides and Toxic Substances
Environmental Protection Agency
A01 M Street, SW
Washington, DC 20460
Attention: Project Officer, Glenn Williams (TS-796) (3 copies)
Contracting Officer, Malcolm P. Huneycutt (MD-33) (1 copy)
TR-835-20
EXPERT REVIEW OF PHARMACOKINETIC DATA:
FORMALDEHYDE
Final Evaluation Report
Prepared Under
Program No. 1415
for
Work Assignment No. 07
Contract No. 68-02-4228
January 2, 1986
-------
£fe Systems, h
DISCLAIMER
This document has not been peer and administratively reviewed within EPA and
is for internal Agency use/distribution only.
-------
ttfe Systems, fa
TABLE OF CONTENTS
PAGE
1.0 INTRODUCTION 1-1
2.0 BACKGROUND 2-1
2.1 Technical 2-1
2.2 Administrative 2-1
3.0 DISCUSSION 3-1
3.1 Distinguishing between Metabolically Incorporated and
Crossllnked CH.O 3-1
3.1.1 Metabolic Incorporations Versus Adduct Formation
of CH 0 . . . 3-1
3.1.2 Crossfinked CH 0 Located Exclusively in the
Interface (IF) DNA 3-2
3.2 Experimental Methodology Limitations 3t-2
3.3 Identity of Labeled Fractions 3-4
3.4 Other Measures of Exposure 3-4
3.5 Nonlinearity for Crosslinked DNA at Low Doses 3-5
3.5.1 Documentation of Nonlinearity for Low Dose
Crosslinked DNA . 3-5
3.5.2 Alternative Explanations for Nonlinearity of Low
Dose Crosslinked DNA 3-6
3.6 Sensitivity of the Study Conclusions to Statistical
Analysis 3-8
3.7 Adequacy of the Measure of Exposure . 3-9
3.8 Utility of the Study in the Quantitative Risk Assessment
of CH20 3-9
4.0 CONCLUSIONS AND RECOMMENDATIONS 4-1
continued-
-------
JCife Systems, Jt
Table of Contents - continued
APPENDIX PAGE
1 Documents Provided to Reviewers Al-1
2 Additional References Relating to Expert Review of
Phannacokinetic Data: Formaldehyde A2-1
3 Questions/Issues on Formaldehyde to be Addressed by
Expert Panel A3-1
4 Information/Clarification Requested from CUT by Expert
Panel AA-1
5 List of Participants A5-1
6 Agenda A6-1
FIGURE
LIST OF FIGURES
Estimated Slopes for Metabolic Incorporation of
Respiratory AQ-DNA and Olfactory IF-DNA ....
PAGE
3-7
11
-------
ttfc Systems, Jn
1.0
INTRODUCTION
This report provides a summary of the discussions and conclusions from the
*eV £ mf,r 8 c°nducted on December 2 and 3, 1985 under Work Assignment (WA)
ll' S'o^Xr £leW °1V?arnaC,°klnetlC Data: Formaldehyde" of Collet
No. 68-02-4228. The preliminary draft of this report was produced on-site by
r K 5 fartlfXMt« to meet the U.S. Environmental Protection Agency
1-1
-------
£ift'Systems. Jt
2.0 BACKGROUND
2.1 Technical
The Environmental Health Committee of EPA's Science Advisory Board (SAB)
reviewed the draft report "Preliminary Assessment of Health Risk to Garment
Workers and Certain Hotre Residents from Exposure to Formaldehyde," dated
May 31, 1985, prepared by the Office of Toxic Substances (OTS). One outcome
of the review is the committee's finding that the formaldehyde (CH.O)
assessment "will not be scientifically adequate without an analysis of the
pharmacokinetlc information and appropriate modification (of the assessment)
based on this analysis." The committee is concerned with both the broad
question: "How can one realistically begin to incorporate relevant kinetic
information into quantitative cancer risk assessments?" and the specific
issue of whether the pharmacokinetic data published In a study by Casanova-
Schmitz et al. (1984) can be used in EPA's assessment of CH 0.
The OTS agrees that appropriate pharmacokinetlc data should be presented and
considered in the risk assessment; however, in its draft risk assessment of
CH.O, OTS agreed with an analysis performed by Cohn et al. (1985) that the use
of the Casanova-Schtnitz data would be premature. OTS found that the Casanova-
Schmltz data do not support a modification of the risk estimated for CH.O^and,
thus, the OTS did not utilize the data either qualitatively or quantitatively
in its risk assessment. However, to address the concern of the Environmental
Health Committee, OTS desired an independent, objective, expert analysis of
the issue so that it may reconsider the appropriateness of using the data in
its risk assessment.
Consequently, OTF authorized assembly of a team of expert scientists to
conduct an evaluation of the Casanova-Schmitz study and any relevant underlying
data developed in the study and to prepare an expert report that answers the
question whether the study provides data that should/could be used in the CH.O
risk assessment.
To accomplish this effort each member of the team of expert scientists was
initially provided the documents listed in Appendix 1. Additional references,
listed in Appendix 2 were provided subsequently to each expert. The experts
were asked to evaluate independently the documents and determine the extent to
which the pharmacokinetlc data are appropriate for use in the CH.O cancer risk
assessment. In conducting their evaluation, the experts were asRed to consider
specifically each of the questions presented in Appendix 3 insofar as their
expertise allows. They were also asked to identify any underlying data (e.g.,
lab books, data tables, etc.) desired from the Chemical Industry Institute of
Toxicology (CUT) which would be of assistance in addressing the specific
questions. The list of information/clarification requested from CUT by the
experts is presented in Appendix 4.
2.2 Administrative
Upon receipt of the VA from EPA, efforts were initiated to assemble a seven-
member team of experts in metabolism, DNA adducts and statistics. The review
2-1
-------
JCifi'Systems.
tear members arc identified on the List of Participants at Appendix 5.
Or. Lemone Yielding agreed to serve as Che Team Leader and Review Meeting
Chairperson. As such, Dr. Yielding served as the principal author/editor of
the technical portions of this report. Other individuals representing EPA and
ICAIR, Life Systems, Inc., at the review meeting are also listed in Appendix 5,
The review meeting was conducted on December 2 and 3, 1985 at the Sheraton
University Center, Durham, NC. The Agenda prepared for the meeting is
provided at Appendix 6. Although the Agenda was prepared for a three-day
meeting, all Agenda items were completed at an accelerated pace and the
meeting only required two days. On-site clerical facilities were provided
during the meeting to prepare the draft report.
2-2
-------
JCifc Systems. Jk
3.0 DISCUSSION
The following provides a summary of the discussions and evaluations conducted
at the review meeting regarding the lists of questions/issues presented in
Appendix 3 and Appendix 4.
Prior to the review meeting, written answers, statements and/or opinions were
prepared by the expert reviewers on each of the questions/issues presented in
Appendix 3. These responses were provided to EPA immediately following the
meeting. They are not provided in this report because they have been
superceded by the consensus opinions developed at the meeting and provided
below.
The representatives from CUT, listed in Appendix 5, participated in informal
discussion with the expert reviewers. This discussion was in response to the
expert's requests for information/clarification listed in Appendix 4 and
Included an expanded question and answer session during which further details
were provided on CIIT's prior, ongoing and planned studies of CH.O.
3.1 Distinguishing between Metabollcally Incorporated and Crossllnked
CH,0 i
1 A^
Questions/issues Nos. I and 3 of Appendix 3 were addressed together and are
discussed below.
3.1.1 Metabolic Incorporations Versus Adduct Formation or Crossllnked
CH2£
The experimental evidence for metabolic incorporation of,crosslinked CH.O is
largely based upon the relative incorporation of H or C-CH.O into DNA of
respiratory epithelium. The evidence presented in the written documentation
was suggestive bur not definitive in regard to this assumption. Additional
evidence, presented at the interview with the CUT staff, Indicated that the
H/ C ratio of the purine deoxyribonucleosides Isolated from aqueous DNA by
high performance liquid chromatography (HPLC) was in accordance with metabolic
incorporation and furthermore, no shifts in pattern suggesting adduct forma-
tion were observed. It was not clear whether or not these types of experi-
ments had been performed at all CH.O levels. Interaction of CH.O with
tetrahydrofollc acid (THFA) could yield N5-10 methylene THFA directly, or
following oxidation to formate could yield N-10 formyl THFA. The occurrence
of these reactions, the equilibration of the various THFA derivatives and the
reaction of the THFA derivatives with various one carbon acceptors could
markedly influence the DNA H/ C ratio in metabolically-lncorporated CH.O.
The pool sizes of the nonradioactlve acceptors and.their metabolic Inter-
mediates could also markedly influence the DNA H/ C ratio due to metabolic
CH.O. The relative reaction rates and pool sizes are likely to vary under
different conditions. Thus, the interpretation of the H/ C ratio is very
complex. This complexity necessitates the isolation of the DNA bases (or
deoxyribonucleosides) and amino acids and the comparison of the H/ C
profiles with authentic standards under conditions which separate any adducts
3-1
-------
fife Systems, h
from the standards. Since It is not clear that this was performed at all CH 0
doses, doubt remains as to the assumption which formed the basis of
distinguishing metabollcally Incorporated and crosslinked (or adducted) CH.O.
3.1.2 Crosslinked CH^O located Exclusively in the Interface (IF) DNA
It is stated by Casanova-Schmltz et al. that all the crosslinked CH.O was
present in the IF rather than the aqueous (AQ) ONA. This assertion was not
borne out by the written or oral documentation. Several control experiments
which would have solidified the experimental basis for this assertion were not
performed. The relative efficiency of extraction of the DNA from respiratory
epithelium under conditions of CH.O dosing (at various levels) should have
been determined. That is, what Is the recovery of the total DNA from the
respiratory epithelium by the phenol procedure? Furthermore, the specific
distribution of crosslinked DNA-protein In the IF-DNA fraction should have
been affirmed. There is no doubt that at least some of the crosslinked CH.O
(DNA-protein) is found in the IF-DNA fraction. Whether all of this crosslinked
fraction was in fact initially extracted from the tissue or whether some might
be found in the AQ-DNA are questions which have not have been satisfactorily
addressed. These experiments could have been performed by adding radio-
actively-labeled DNA (or crosslinked DNA-protein) to a tissue homogenate and
following its distribution into the various fractions of the phenol procedure,
i.e., a standard recovery experiment. Consequently, we believe that sufficient
detail for characterization of the IF-DNA fraction has not been provided.
This detail is absolutely necessary in order to validate the use of IF-DNA as
a measure of crosslinked CH.O, and therefore as a biological dosimeter.
3.2 Experimental Methodology Limitations
The inhalation methodology used to administer CH.O was appropriate and the
control and monitoring of CH.O concentration as well as the isotopic composi-
tion of the gas mixtures employed were adequate. Infrared monitoring was done
continuously during exposure and the instruments and methods crosschecked with
other instruments and other methods. Variations in CH.O concentrations were
small compared to other variables and are not likely to have much impact on
the experimental results. These factors are not considered to represent an
important source of error in the experimental protocol. In considering the
interpretation of the data in relation to human risk assessment, it should be
borne in mind that Inhalation by rats is restricted to the nose. In humans,
inhalation can be expected to occur through both the nose and the mouth. The
implications of this distinction with respect to human risk are unclear but
always represent a source of uncertainty when rodents are employed as hucan
surrogates in inhalation studies.
The DNA extraction procedure employed requires further validation since It is
of central importance in distinguishing between the aqueous and Interfaclal
DNA fractions on which the measurements of metabolic Incorporation and crosslink-
Ing depend. There is no indication of the percentage of the total DNA recovered
by this procedure, and it is likely that the amount of DNA associated with the
interfacial fraction will vary to some extent with the extraction conditions
employed (e.g. ionic strength, temperature, etc.). A standard recovery experi-
3-2
-------
JCifc Systems. JH
ment as outlined in 3.1.2 should have been conducted. The adequacy of the
extraction procedure can be assessed only after a more complete characteriza-
tion of the nature of the DNA occurring In the Interfacial fraction. Small
variations In the extraction conditions could constitute an Important source of
experimental error and could exert a profound Influence on the H/ C ratios
obtained. It would be reassuring If, after protelnase treatment and hydro-
xyapatite chromatography, a mild acid hydrolysis (with or without added carrier
CH.O) could be shown to release radio-labeled CH.O from this material. This
would give positive evidence for the existence or chemically-incorporated CH.O
in the IF-DNA fraction.
Exposures to labeled CH.O were routinely conducted over a period of six hours
with rats that had been preexposed to the same concentration of the gas for
six hours the previous day. In view of the established temporal changes in
cell proliferation as well as the replacement of respiratory epithelial cells
by squamous cells during chronic exposure, there is some question whether the
results of the acute labeling studies will accurately reflect events occurring
during longer-term exposures.
This is particularly important when it is considered that squamous cell
carcinoma does not develop until 11 or 12 months into the chronic study and
that a large percentage of DNA protein crosslinks are subject to relatively
rapid repair. CUT investigators argue that the short-term studies are the
most appropriate models for human exposure since in humans there is no evidence
of the marked changes in epithelial cell structure that are observed during the
chronic rat studies. On the other hand, it is not really clear as to whether
or to what extent CH-O-DNA interactions differ during the course of chronic
exposure.
Thus, the short-term conditions employed to evaluate CH.O binding may or may
not reflect those occurring during chronic exposures, and there is con-
siderable uncertainty in relating the acute binding data directly with the
carcinogenic lesions occurring as a result of chronic exposure.
While clearly it is not possible to conduct binding studies throughout the
entire period of the chronic test, it would be useful for comparative purposes
to have data from animals exposed to CH.O for a longer period of time. The
primary uncertainty associated with the data is that they represent an acute
model of a chronic lesion. The extent to which this model is valid remains
to be determined.
Information provided by CUT scientists at the meeting addressed satisfactorily
initial questions concerning methodology for determination of H/ C ratios.
Samples of raw data were provided which clearly showed very good counting
statistics. The problems, which could potentially arise from quenching, were
minimized and controlled through (1) routine use of external standardization
(built into the scintillation counter), (2) routine counting of quenched and
unquenched standards with samples, and (3) using a relatively constant ratio
of scintillation fluid.to.the aqueous samples. For standardization between
experiments, observed H/ C ratios were normalized to the H/ C ratio of the
gas phase to which the animals were exposed.
3-3
-------
£i(c Systems, k
There are numerous points at Vn*(b tritium kinetic isotopr effects could
influence the disposition of H/ C dual-labeled CH 0. For example, there
could be a large primary Isotope effect on the enzyraic oxidation of CH 0 to
formate. Smaller secondary isotope effects would influence the addition of
nucleophiles to monomeric CH.O and the reactions by which formate equivalents
are incorporated into purine bases. It would, in fact, be amazing if these
kinds of isotope effects did not enter into the results. The problems arise
because of the complexity of the overall disposition of CH.O (see
Section 3.1.1). This affords the opportunity that under different conditions
of administration the relative importance of the various steps will vary.
Since the overall (observed) isotope effect will be a composite (a weighted
average) of the Isotope effects on these individual steps, there is great
opportunity for quantitative variation among the. individual steps even in the
face of an apparently constant value for the H/ C ratio observed in a
particular fraction such as AQ-DNA. This is not to suggest that there are
problems of this sort with the data presented* <"?Jy that there is reason for
concern that there might be. Measurement of H/ C ratios found in specific
bases obtained by hydrolyzing AQ-DNA and 1F-DNA after both high and low doses
of CH.O would help clarify this situation.
3.3 Identity of Labeled Fractions
i
This issue/question is addressed under Section 3.1 and the last part of
Section 3.2.
3.4 Other Measures of Exposure
Questions/issues Nos. 4 and 7 of Appendix 3 were addressed together and are
dlscuFFed below.
While a measure of the effective concentration of a chemical at its target
site (i.e., the delivered dose) would be preferable to the use of administered
dose for purposes of risk assessment, it is questionable whether the DNA-
binding data generated by Casanova-Schmltz et al. provide a validated measure
of CH 0 concentration in the nasal epithelium of exposed rats. Since the
relationship between DNA binding and carcinogenicity have not yet been
established for CH.O, the observations on binding may be of some mechanistic
significance. However, while DKA-binding may constitute a satisfactory
measure of target site (delivered) dose, its measurement by means of
dual- labeled CH.O incorporation seems unnecessarily sophisticated and complex.
What is really needed- as a biochemical dosimeter is some measure of the
chemical interaction of CH.O with intracellular macror.olecules other than
through metabolic incorporation into nucleotides and/or amino acids.
DNA-protein crosslinks, if they could be shown unambiguously to involve
methylene links originating from administered CH.O, would be a valid measure
of this. Alternatively, since the jln vitro experiments show that proteins are
far more reactive toward CH.O than nucleic acids, and since the metabolic
Incorporation of CH.O equivalents into proteins is likely to be much less than
its metabolic incorporation into nucleic acids, the measurement of CH.O
covalently bound to Intracellular proteins could provide a simpler index of
3-4
-------
fife Systems, h
exposure of the cell to CH«0. Furthermore, validation of these assumptions
could easily be accomplished.
While the data provided by the Casanova-Schmitz et al. study may provide some
measure of delivered dose, It Is not yet clear whether the data are a great
deal more useful than measures of administered dose for the purposes of risk
assessment. In the absence of DNA binding data obtained following longer
periods of exposure, there is no Justification for assuming that the short-
term data bear any relationship to target-site concentrations likely to be
encountered throughout the two-year bioassay.
3.5 Nonlinearlty for Crosslinked DNA at Low Doses
3.5.1 Documentation of Nonlinearlty for Low Dose Crosslinked DNA
Casanova-Schmitz et al. calculate the amount of covalent binding from
equations Nos. 8 and 9 in their Appendix 2. The issue of whether the result
actually represents the extent of crosslinked DNA is addressed elsewhere in
this report and will not be discussed further here. The authors have assessed
the nonlinearity of the dependence of crosslinking on administered dose in two
ways.
j
1. In the discussion section in Casanova-Schnitz et al. there Is a
comparison between:
a. The actual values calculated from data at 2 ppm.
b. The value predicted by interpolating between the results for
6 ppTr and the origin. The observed values are 0.022 +/-
0.005 nmol/mg (mean +/- 1 ISE), while the interpolated values
are stated to be 0.078 +/- 0.013 (one third of the 6 ppm values,
0.233 +/- 0.023; we note that simple division gives 0.078 +/-
0.0078). The conflict between observed and predicted values is
clear.
2. In Appendix 3 of Casanova-Schritz et al., a more systematic test of
proportionality is described. The basic idea is to convert the
concentration at each administered dose into an estimated slope by
dividing by the dose. If the response were truly proportional to
dose, these estimated slopes would all be the same. Therefore, the
hypothesis may be tested by examining the consistency of the slopes.
Since the standard errors of the estimated slopes are more or less
constant, and in particular show no systematic tendency either to
increase or decrease with Increasing dose, the statistical method of
one-way analysis of variance is appropriate and leads to rejection
of the null hypothesis of proportionality.
It has been suggested by Cohn (1984) and Cohn et al. (1985) that olfactory
IF-DKA may give a better indication of metabolic incorporation than does
respiratory AQ-DNA, and that when calculated In this alternative way, the
extent of crosslinking becomes proportional to dose. This suggestion has been
3-5
-------
£ife Systems, JH
disputed by the Casanova-Schmltz et al. authors. However, they argue in their
rebuttal that even when calculated In this alternative way, there is still
strong evidence of nonllnearity. The comparison between results at 2 pptn and
6 ppo is no longer sharp, and the one-way analysis of variance yields only
marginally significant results. However, the linear regression against dose
is statistically significant, and the authors correctly point out that this is
a more powerful test against alternative hypotheses of the type expected here.
Figure 1 provides a graphical assessment of the proportionality hypothesis.
It shows the estimated slopes calculated in both ways with standard errors
attached, plotted against dose. The left-hand bar of each pair uses respiratory
AQ-DNA as the baseline, while the right-hand bar is based on olfactory IF DNA.
Although the nature of the nonconsistency is different for the two alternatives,
it is strong in both. From a purely statistical point of view, the weakness
of the use of olfactory IF-DNA is that weaker pairing of the data leads to
generally larger standard errors, and consequently lower power to detect
nonconsistency (nonlinearity of amount of crosslinking).
To summarize, in contrast to the claims of Cohn (1984) and Cohn et al. (1985),
we find that the nonlinearity of "crosslinked DNA" as a function of administered
dose is adequately documented by Casanova-Schaitz et al.
3.5.2 Alternative Explanations for Konlinearity of Low Dose
Crosslinked DNA
Each determination of crosslinked DNA is based on three replications with each
replicate based on material from four animals. With such small numbers of
measurements, it is always possible that spurious results may be obtained.
The calculation of .significance levels is the statistician's way of trying to
evaluate the weight of evidence in small samples and to avoid being misled by
spurious indications. The Casanova-Schmitz et al. authors have been careful
to calculate these. However, additional data, especially below 6 ppm, would
add considerable substance to the results. There are other mechanisms that
might lead to apparent nonlinearity. For instance, suppose that there were a
small but constant loss in the measurement of IF-DNA. This would be increasingly
important at low doses, and hence, would Induce an apparent threshold in the
amount of crosslinked DNA. As indicated in Section 3.2, the efficiency of
extraction of DNA and the verification of distribution are of prime importance
in excluding possibilities such as these.
The authors have suggested that physiological and biochemical defense mechanisms
could become less efficient with increasing CH.O concentrations. We agree
that the processes of tnucociliary clearance ana DNA repair could become
saturated as the CH.O concentration increases, and DNA-protein crosslink
formation could therefore increase disproportionately.
3 14
Furthermore, the disproportionate increase in H/ C ratio with increase in
CH 0 concentration might be due to artifactual disturbances in the H/ C
ratio rather than a true increase in crosslinked DNA-protein in the IF frac-
tion. The effect of CH.O is somewhat complicated, since CH.O is known to
inhibit DNA synthesis, yet cause an increase in cell turnover in a small
3-6
-------
jCife'Systems, M
0.05-
0.04 -
0.03-
0.02-
0.01-
0 «4
A Respiratory AQ-DNA*
• Olfactory IF-DNA**
T
2
]
6
10
15
Administered Level of Formaldehyde, ppm
(a) Casanova-Schmitz, M., Starr, T.B., and Heck, H. D'A. (1984) Differentiation
Between Metabolic Incorporation and Covalent Binding on the Labeling of
Macromolecules in the Rat Nasal Mucpsa and Bone Marrow Inhaled ("Q- and
(3H) CH20. Toxicology and Applied Pharmacology 76,26-44.
(b) Memorandum to Peter W. Preuss from Murray S. Conn concerning "Health
Sciences comments in response to the Environmental Protection Agency's
request for information regarding CH2O..!' dated July 16,1984.
FlttJRE 1 ESTIMATED SLOPES FOR METABOLIC INCORPORATION
OF RESPIRATORY AQ-DNA AND OLFACTORY IF-DNA
3-7
-------
JCifcSystenis.Jh
percentage of cells. Inhibition of DNA synthesis would decrease metabolic
incorporation but might also decrease sites for adduct formation if these are
restricted to the replication fork. More experimentation ia thus necessary to
determine how Inhibition of DNA synthesis and/or increase in cell turnover
influence the H/ C ratio in IF-DNA.
Cell death might also be involved in the nonproportional response in the
low dose range. Thus, if cell death occurs, nucleic acids are released and
degraded. The subsequent increase in nucleotldes, nucleosides, purines and
pyrimidines could inhibit de_ novo nucleotide synthesis from CH.O by feedback
regulation as well as increase the deoxynucleotide pools, thus, diluting out
radioactivity Incorporated into DNA. A decrease in DNA synthesis from labeled
CH.O. in.the absence of an effect on adduct formation, would tend to increase
the H/ C ratio disproportionately at higher CH 0 concentrations.
3.6 Sensitivity of the Study Conclusions to Statistical Analysis
The statistical methods for the comparison of AQ- and IF-DNA in terms of
incorporated concentrations of ( C) CH.O were described as a two-way analysis
of variance followed by paired t-tests For Individual concentrations. The two
factors in the analysis of variance were concentrations of CH.O and type of
DNA: IF and AQ. It is not clear If the pairing of IF- and AQ-DNA determina-
tions were taken into account In the two-way analysis of variance. Moreover,
the lack of homogeneity of variance between the low and high concentrations
was apparently not taken into account in the analysis; neither were ordered
alternatives over the concentrations when the two-way analysis of variance led
to statistically significant tests. Thus, more powerful procedures could have
been employed to examine the differences In concentrations of ( C) CH.O
between IF- and AQ-DNA over CH 0 concentrations.
An examination of the data on respiratory DNA reveals that there Is a statis-
tically verifiable increase in the concentration of bound ( C) CH.O per mg
DNA over CH.O concentration as concluded in the paper. These comments also
apply to the comparisons of H/ C ratios between IF- and AQ-DNA as a function
of CH 0 concentrations. Thus, the overall conclusions on the comparisons of
IF- and AQ-DNA as a function of CH.O concentrations hold even though the
statistical analysis could be strengthened.
In their comparisons of IF- and AQ-DNA with paired t-tests at separate CH.O
concentrations, Casanova-Schnitz et al. concluded that the bound ( C) CH.O
concentrations did not differ significantly at 0.3 and 2 ppm. The power of
these paired t-tests is low for these CH.O concentrations because of the small
sample sizes at individual concentrations relative to the coefficient of
variability. Although this is not critical with respect to the pattern of the
IF- and AQ-DNA data over the CH.O concentrations used in the experiment, it
does limit the extent to which Inferences can be made about the responses to
low concentrations for purposes of identifying no-response levels or making
lov-dose extrapolations for risk assessment.
3-8
-------
ttfe Systems. Jt
3.7 Adequacy of the Measure of Exposure
This question/Issue was fully addressed along with No. 4 in Section 3.4 above.
3.8 Utility of the Study in the Quantitative Risk Assessment of
CH2Q
The panel recognized this study as an important step toward attempting to
assess the intracellular dose delivery of externally applieded CH_0. These
efforts should be continued toward the ultimate goal of improving the assess-
ment of risk. At its present level of development and validation, however, the
study does not represent an adequate basis for quantitative risk assessment.
First, the problem of proper validation of the experimental methodologies must
be accomplished to assure that CH.O-DNA-proteln complexes are identified
properly, and that the experimental assumptions are valid. The evidence is not
sufficiently strong at this time to reject the linear dose extrapolation model.
Second, the selection of a single intracellular target is complicated by the
nature of binding processes with DNA and could be augmented appropriately by
the additional analysis of binding to Intracellular proteins. Third, and
perhaps most important, the selection of the acute model may not be entirely
appropriate since it is the chronic dosimetry that is most relevant to risk
assessment. The factors which account for nonlinearity of dose delivery may
veil vary considerably (in either direction) as a result of chronic treatment.
This study is an important first step toward the introduction of Intracellular
dosimetry Into the risk assessment process. The continuation and extension of
these investigations should be encouraged.
3-9
-------
fife Systems, Jn
4.0 'CONCLUSIONS AND RECOMMENDATIONS
The following summarizes the conclusions and recommendations developed by the
review team during the December 2 and 3* 1985 meeting. These conclusions and
recommendations have undergone final review by each of the experts participating
in the meeting.
1. Some doubt still remains as to the validity of the assumptions which
form the basis for distinguishing metabolically incorporated and
crosslinked (or adducted) CH20, i.e., H/ C in DNA.
2. Experimental methods and controls were adequate with respect to
monitoring the CH.O administration and analysis of dual-labeled
materials. However, the chloroform/iso-amylaIcohol/phenol extrac-
tion for DNA and DNA crosslinked to proteins was not validated in
terms of the identities of materials separated nor the overall
efficiency and consistency of extraction. The occurrence of under-
lying variability Incorporation due to kinetic isotope effects on
the disposition of tritiated CH.O can neither be assessed nor
discounted. t
3. Sufficient documentation is still unavailable to state unequlvocably
that all the crosslinked DNA-proteln complexes occur in the IF-DNA
fraction.
4. There remains a need for an effective biochemical dosimeter to
measure the dose of CH.O delivered to the cells of the nasal epithe-
lium. The data provided by Casanova-Schnitz et al. are not
considered a sufficiently well-validated measure of this parameter.
5. The nonproportionality of the calculated concentration of bound
C (CH.O)-DNA as a function of the administered dose is documented
adequately. Vhether the nonproportlonality truly reflects crosslink
formation or is due to the small sample size, to a constant loss in
the recovery of IF-DNA, or to artifactual disturbances In the H/ C
ratio remains to be elucidated.
14
6. The Increase in concentration of bound C with the concentration of
CH.O.is veil documented, as Is the Increase in the difference In the
H/ C ratio between IF- and AQ-DNA. The power of separate compari-
sons for the 0.3 and 2 ppm doses Is low because of small sample size
relative to the coefficient of variation. This limits the potential
for inferences about no-response levels and lov-dose extrapolations.
7. The study of Casanova-Schmitz et al. is an important first step
toward quantitative assessment of the intrace'llular level of CH.O in
the nasal mucosa of the rat following inhalation exposure. At Its
present level of validation, however, It does not provide a basis
for such quantitation. Furthermore, the selection of an acute study
model may not be appropriate to the assesscent of chronic toxicity.
4-1
-------
£tf< Systems, J>
APPENDIX 1
DOCUMENTS PROVIDED TO REVIEWERS
1. Preliminary Assessment of Health Risks to Garment Workers and Certain
Home Residents Exposure to Formaldehyde. EPA Draft Report. May 31, 1985.
2. Casanova-Schmitz, M., Starr, T.B., and Heck, H. D'A. (1984) Differentia-
tion Between Metabolic Incorporation and Covalent Binding on the Labeling
of Macromolecules in the Rat Nasal Mucosa and Bone Marrow Inhaled ( C)-
and ( H) CH?0. Toxicology and Applied Pharmacology 76, 26-44.
3. Cohn, M.S., DiCarlo, F.J., and Turturro, A. (1985) Letter to the Editor.
Toxicology and Applied Pharmacology. 77, 363-364.
4. . (1985) Letter to the Editor. Toxicology and
Applied Pharmacology. 77, 365-368.
5. . (1985) Letter to the Editor. Toxicology and
Applied Pharmacology. 77, 358-361.
6. Selected comments pertaining to the use of the CIII "effective dose"
experiment. . .
7. Memorandum to Peter W. Preuss from Murray S. Cohn concerning "Health
Sciences comments in response to the Environmental Protection Agency's
request for information regarding CH.O. . ." dated July 16, 1984.
Al-1
-------
£ife Systems. Jn
APPENDIX 2
ADDITIONAL REFERENCES RELATING TO EXPERT REVIEW OF
PHARMACOKINETIC DATA: FORMALDEHYDE
1. Starr TB, Buck RD. 1984. The Importance of delivered dose In estimating
low-dose cancer risk from inhalation exposure to formaldehyde. Fund.
Appl. Tox. 4:740-753.
2. Comments to the EPA Science Advisory Board by CUT scientists regarding
the EPA draft entitled "Preliminary Assessment of Health Risks to Garment
Workers and Certain Home Residents from Exposure to Formaldehyde."
June 21, 1985.
3. Comments to the EPA Science Advisory Board by Dr. James A. Svenberg
regarding the EPA draft entitled "Preliminary Assessment of Health Risks
to Garment Workers and Certain Home Residents from Exposure to Formaldehyde."
July 9, 1985.
4. Comments to the EPA Administrator by the Environmental Health Committee of
EPA's Science Advisory Board regarding the EPA draft entitled "Preliminary
Assessment of Health Risks to Garment Workers and Certain Home Residents
from Exposure to Formaldehyde." Undated.
A2-1
-------
&fe Systems, Jn
APPENDIX 3
QUESTIONS/ISSUES ON FORMALDEHYDE TO BE ADDRESSED BY
EXPERT PANEL
1. Are Che assumptions which form the basis for distinguishing between
metabolically incorporated and crosslinked formaldehyde adequately
supported?
2. The appropriateness or limitations of the experimental methodology
used.
3. What do the measurements taken in the Casanova-Schmitz study represent-
i.e., is there ambiguity in the identity of the various labeled fractions?
Do the data establish that the IF fraction consist of crosslinked DNA?
4. Are there any other data in the study that could be used as a measure of
exposure in addition to the crosslinked DNA?
5. Is the nonlinearity for crosslinked DNA at low doses adequately docu-
mented? Are there alternative explanations for these observations? ;
6. What is the sensitivity of the conclusions of the study to both experi-
mental error and the statistical treatment of the data?
7. Does the study give a better measure of exposure than the "applied dose"
for the second day post exposure; and If so, does It also give a better
measure of the dose durinp the two-year bionssay?
8. Conclusions the experts can draw concerning the utility of the study
(and any underlying data) in the quantitative risk assessment of
formaldehyde.
A3-1
-------
£ifc Systems, Jut
APPENDIX 4
INFORMATION/CLARIFICATION REQUESTED FROM CUT
BY EXPERT PANEL
1. Would like to see Information on the methodology and equipment used for
scintillation counting. Needs information on quench correction and the
methods used to calculate DPMs.
Would also like to see an example of the raw data used to calculate DPM.
2. Would like to review raw data used to perform two-way analysis of
variance between aqueous phase and.interfacial DNA with respect to their
incorporation concentrations of ( C) CH_0 equivalents or
their H/ C ratios at different concentrations of CH.O.
^ ~,, « —*„_, — with respect to
3. Request copies of the following references:
a. Svenberg, J. A., Gross, E. A., Martin, J., and Popp, J.A. (1983a).
Mechanisms of formaldehyde toxlclty. In Formaldehyde Toxicity
(J. E. Gibson, ed.)» pp. 132-147. Hemisphere, Washington, D.C.
b. Swenberg, J. A., Gross, E. A., Randall, H. W., and Barrow, C. S.
(1983b). The effect of formaldehyde exposure on cytotoxicity and
cell proliferation. In Formaldehydet Toxicology, Epidemiology
and Mechanisms (J. J. Clary, J. E. Gibson, and R. S. Waritz, eds.),
pp. 225-236. Dekker. New York,
AA-1
-------
JCifc Systems, Jnc.
APPENT1X 5
TR-835-19
LIST OF PARTICIPANTS
Expert Review of Pharmacokinetic Data:
Formaldehyde
Participants
Dr. Edward Bresnlck
Eppley Institute for Research
in Cancer
University of Nebraska Medical
Center
Omaha, NE 68105
(402) 559-4238
Dr. Peter Bloomfield
North Carolina State University
518 Cox Hall
Raleigh, NC 27695
(919) 737-2541
Dr. Richard Cornell
Department of Biostatistics
School of Public Health
University of Michigan
Ann Arbor, MI 48109
(313) 764-5450
Dr. Helen Evans
Department of Radiology
Case Western Reserve University
2065 Adelbert Road
Cleveland, OH 44106
(216) 844-3530
Dr. Robert P. Hanzlik
Department' of Medicinal Chemistry
University of Kansas
Lawrence, KS 66045
(913) 864-3750
Dr. Christopher F. Wilkinson
Cornell University
N202 MVR
Ithaca, NY 14853
(607) 256-8112
Dr. Lenone Yielding***
Department of Anatomy
University of South Alabama
2042 Medical Sciences Building
Mobile. AL 36688
(205) 460-6490
EPA Staff
Dr. William Farland
Health and Environmental Revlev
Division
Office of Toxic Substances
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
(202) 382-4241
CUT Representative(s)
Dr. Mercedes Casanova-Schmitz
Dr. James Gibson
Dr. Henry D'A Heck
Dr. Thomas B. Starr
Chemical Industry Inst. of Toxicology
P.O. Box 12137
Research Triangle Park, NC 27709
(919) 541-3440
ICAIR, Life Systems, Inc.
Dr. John P. Glennon, Program Manager
Mr. Daniel Mecklcy, Task Manager
24755 Highpoint Road
Cleveland, OH 44122
(216) 464-3291
(a) Chairperson.
A5-1
-------
£ife Systems, Jut.
Time
APPENDIX 6
TR-835-18
AGENDA
EXPERT REVIEW OF PHARMACOKINETIC DATA: FORMALDEHYDE
December 2-4, 1985
Chamber C, Greenbrier Ballroom
Sheraton University Center
Durham, NC
Meeting Chairperson: Dr. Lemone Yielding
Agenda Item Individual
Monday, December 2, 1985
(a)
9:00 a.m. Informal Technical Discussions
12:00 noon Break
1:00 p.m. Welcome
1. Administrative Announcements
2. Summary of EPA's Needs
1:45 p.m. Meeting Objectives
2:00 p.m. Chairman's Opening Comments
2:30 p.m. . Input From CUT
4:30 p.m. Finalize List of Questions and Issues
5:30 p.m. Break
7:00 p.m. Discussion of Questions and Issues
1. Discussion
2. Consensus
3. Assignment of Draft Report Authors
9:00 p.m. Adjourn for Day
(b)
D. Meckley
V. Farland
J. Glennon
L. Yielding
Dr. Heck
L. Yielding
L. Yielding
contlnued-
(a) Optional for those Individuals arriving December 1, 1985.
(b) Selected authors of Draft Report sections may adjourn to prepare rough
draft or entire meeting may adjourn to prepare Draft Report sections as
determined by the participants.
A6-1
-------
£ife Systems. JM
Ap.cnda - continued
Time
Agenda Item
Tuesday, December 3, 1985
8:00 a.m. Administrative Announcements
8:10 a.m. Chairman's Comments
8:20 a.m. Discussion of Questions and Issues
1. Discussion
2. Consensus
3. Assignment of Draft Report Authors
12:00 noon Break
1:00 p.m. Discussion of Questions and Issues - continued
3:00 p.n. Meeting Status Summary
1. Questions and Issues Resolved
2. Remaining Action Items
3. Revision of Agenda/Schedule
4:00 p.n. Discussion of Questions and. Issues - continued
5:00 p.m. Adjourn for Day
Wednesday, December 4, 1985
8:00 a.m. Administrative Announcements
8:10 a.m. Draft Report Status
8:20 a.m. Preparation of Draft Report
1. Complete Action Items
2. Review/Discuss Draft Report Sections
3. Revise Draft Report Sections
12:00 noon Break
1:00 p.m. Preparation of Draft Report - continued
4:00 p.m. Final Review of Draft Report Status
5:00 p.n. Adjourn
Individual
D. Meckley
L. Yielding
L. Yielding
L. Yielding
L. Yielding
J. Glennon
L. Yielding
D. Meckley
L. Yielding
L. Yielding
L. Yielding
L. Yielding
J. Glennon
A6-2
-------
m
APPENDIX 1: EXPERT PANEL REPORT ON HCHO
PHARMACOKINETIC DATA AND CUT RESPONSE
Cinci
.
nnati ,
-------
Chemical Industry Institute of Toxicology
President. Robert A. Neal. PhD.
Vice President. Director of Research. James E. Gibson. PhD.
Vice President. Administration and Secretary. Donald A. Hart. Ed.D.
February 4, 1986
P. O. Box 12137
Research Triangle Park,
North Carolina 27709
(919)541-2070
Dr. William H. Farland
Deputy Director
Health and Environmental
Review Division
Office of Pesticides and
Toxic Substances
U. S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Dear Dr. Farland:
Enclosed are detailed comments on the document "Expert
Review of Pharmacokinetic Data: Formaldehyde" which was
authored at the meeting in Research Triangle Park on
December 2-4. I have also sent copies of these comments to
the other members of the committee inviting their individual
or collective comments. We would also welcome any comments
you personally might have concerning points raised in this
critique.
Sincerely,
Robert A. Neal
President
RAN:ewb
Enclosure
-------
COMMENTS ON THE FINAL REPORT OF THE PANEL REVIEWING THE CUT PHARMACOKINETIC
DATA ON FORMALDEHYDE
M. Casanova, T. B. Starr, and H. d'A. Heck
We have carefully examined the final report of the Panel reviewing the
CUT pharmacokinetic data on formaldehyde. We find many of their concerns to
be without merit, and we disagree strongly with their conclusions. A de-
tailed justification for this assessment of the Panel's report is given below.
3.1.1. Metabolic Incorporation versus Adduet Formation or Cross I inked CH^O
&~
The Panel states that the "interpretation of the 3H/14C ratio of the DMA
due to metabolic incorporation of [ H]- and [ CJCHJ)] is very complex" owing
to the fact that 'relative reaction rates and pool sizes are likely to vary
under different conditions'. Hence, the Panel argues that it is necessary to
isolate the DNA bases by HPLC and to determine the isotope ratios of the
isolated bases. The Panel did not mention that we have already undertaken
such studies, although we informed the Panel of this on December 2, 1985, when
we had our meeting with then. The Panel did not request details of our
studies either at the meeting or subsequent to the meeting, although we
volunteered to provide any information requested. It is surprising to us,
therefore, that they called in their report for HPLC studies to be done.
-1-
-------
For the record, we would like to provide the essential results of our HPLC
analyses of DNA. The DNA samples that were analyzed were obtained from rats
3 14
that had been exposed (6 hr) to 0.3, 2, or 6 ppra of [ H]- and [ C]formalde-
hyde. These were the same samples that had previously been used for determin-
ations of covalently bound CHgO in DNA (Casanova-Schmitz et £L, 1984). Suf-
ficient AQ DNA remained from those samples for analysis by HPLC. However,
only one IF DNA sample remained from the initial experiments, and that sample
was obtained from rats exposed at 6 ppm.
The major UV-absorbing peaks from respiratory mucosal AQ DNA samples elu-
ted at the same positions as authentic purine and pyrimidine deoxyribonucleo-
side standards. Calculation of the base compositions of the DNA samples was
performed after calibration of the UV monitor with nucleoside standards. The
values obtained for the base compositions of the DNA samples agreed well with
one another and with the base compositions reported in the literature; obs.:
deoxyadenosine (dAdo), 28.9 * 0.3 X; deoxycytidine (dCyd), 21.3 * 0.4 X; lit.
values (Shapiro, 1968): dAdo, 28.8 * 0.7 X; dCyd, 20.5 * 0.5 X. This signi-
fies that the base composition of the AQ DNA is the same as that of the total
rat DNA. The nucleosides, deoxyguanosine (dGuo) and thymidine (dThd), were
not completely resolved in the chromatography, hence, their individual base
compositions were not calculated.
As expected (Casanova-Schmitz et aj_., 1984), most of the radioactivity in
the AQ DNA from the respiratory nucosa eluted at the positions of the normal
deoxyribonucleosides, dGuo, dThd, and dAdo, implying that the labeling of the
AQ DNA was primarily caused by normal metabolic incorporation. In three AQ
-2-
-------
DNA samples from rats exposed to 6 ppm of CH-0, the percentage of the total
C that eluted at the positions of the normal nucleosides was 100%, 96%, and
90%, respectively. In two of the three samples, small amounts of radioactiv-
ity eluted prior to the .major peaks, but no radioactivity eluted after dAdo in
•any sample. A Iate-eluting radioactive peak would be expected for 6-hydroxy-
*••
methyl-deoxyadenosine, a postulated adduct of formaldehyde with DNA (BeI and et
al., 1984). If such an adduct were formed, it did not remain in our DNA
samples until the time of analysis.
3 14
The H/ C ratios of the major peaks were consistent with normal metabolic
incorporation. The isotope ratios of the deoxyribonucleosides did not vary
measurably with concentration over the range 0.3 to 6 ppm. We observed that
the 3H/14C ratio of dAdo (0.55) was higher than that of dGuo (0.25). A higher
isotope ratio for dAdo than for dGuo is consistent with the known pathway of
formaldehyde incorporation via tetrahydrofolate into positions 2 and 8 of ino-
sine monophosphate (IMP), the precursor of both GMP and AMP. The conversion
of IMP to GMP involves the introduction of an oxygen atom at position 2 of the
purine ring (with consequent loss of H at this position). The conversion of
3
IMP to AMP does not involve a corresponding loss of H at position 2. Thus,
AMP should have a higher H/ C ratio than GMP, as observed experimentally.
We observed no labeling of dCyd in our chromatograms. This implies that
the labeling of dThd (3H/14C • 0.4) was due only to labeling at the 5-methyl
position, which results from transfer of the nethylene carbon atom from N ,-
N -methyIene-tetrahydrofolate to deoxyuridine monophosphate.
-3-
-------
3 14
The H/ C ratios of the minor peaks seen in two of the three chromato-
grams were low (ranging from 0.3 to 0.4), resembling the ratios seen in the
major peaks. This suggests that the minor peaks of radioactivity were not due
3 14
to covalent adducts, which would be expected to have higher H/ C ratios than
those seen in the normal bases. It is likely that at least some of the minor
»•
f.
radioactive peaks in the AQ DNA samples were due to slight contamination of
the DNA with RNA. This hypothesis is supported by the observation that minor
peaks eluted at positions similar to those obtained using ribonucleoside stan-
dards, adenosine and guanosine. It is also possible that a minor peak resul-
ted from deamination of dAdo to deoxyinosine, which might have occurred if the
alkaline phosphatase used in the hydrolysis of DNA had been contaminated with
adenosine deaminase (Gehrke et a_L, 1982). (No loss of H would occur in this
3 14
reaction, therefore, the H/ C ratio would remain unchanged.) Finally, a mi-
nor peak may have been due to 5-methyl-deoxycytidine (5-MedCyd), a normal con-
stituent of rat DNA accounting for about 1% of the bases (Shapiro, 1968). The
presumed labeling of the 5-MedCyd would be expected to occur via transfer of a
methyl group from S-adenosylmethionine to dCyd in DNA (Kornberg, 1980). This
methyl group could be labeled, since methionine might be synthesized in small
amounts from methyl-tetrahydrofolate, although methionine is usually consi-
dered to be an 'essential' ami no acid.
It should be noted that even if one assumes the most extreme case that 10X
of the radioactivity in the AQ DNA were due to RNA, the error in estimating
the 3H/14C ratio of the AQ DNA would be only IX, due to the similar isotope
ratios of RNA and AQ DNA. Moreover, since RNA has a higher specific activity
than DNA (Casanova-Schmitz et ja_L, 1984), the actual contamination of the DNA
-4-
-------
by RNA would be less than 4% in the most extreme case. Errors of these magni-
tudes are completely negligible in our calculations of covalent binding.
3 14
A single IF DMA sample from rats exposed to 6 ppm of [ H]- and [ CJCH-0
was also analyzed by HPLC. In this sample, one additional peak was seen that
was not present in any AQ DMA sample. This*' peak eluted very early in the
•
chromatogram and had an apparent H/ C ratio > 1.0. Such a peak could con-
ceivably be incompletely digested DNA containing covalent I y bound GO, the
hydrolysis of which was prevented by ' DMA-ami no acid or DNA-peptide cross-
linking. This interpretation is, of course, tentative, and additional studies
are needed to test this hypothesis. As in the AQ DNA, there were no peaks
eluting after dAdo.
It is difficult to unequivocally identify minor peaks seen in HPLC, owing
to the low levels of radioactivity and the small amounts of DNA in the IF DNA
samples. After enzymatic hydrolysis, each of the major metabolicaIly-labeled
nucleoside peaks in the IF DNA contained only about 100 to 200 dpm, and the
total radioactivity in the unidentified early-eluting minor peak in IF DNA
with the high H/ C ratio was only 80 dpm. Thus, owing to the small amount
of DNA obtainable from the rat nasal mucosa, the low level of radioactivity in
the IF DNA that was due to cross-1 inking, and the present lack of information
concerning the detailed structures of the DNA-protein cross-links, we doubt
that much more information than we have already obtained would result from
continuing or extending the HPLC studies, despite the recommendation of the
Panel. At least the qua Iitative outcome of such studies has already emerged
from our work.
-5-
-------
It should be emphasized that the results of our HPLC analyses of DMA sam-
ples collected from exposed rats are in full agreement with our conclusions
that AQ DNA does not contain covalently bound ChLO, j_..e., that the labeling of
AH DNA is due to metabolic incorporation, whereas the labeling of IF DNA is
caused both by metabolic incorporation and covalent binding. We are satisfied
*•
that the concerns of the Panel concerning the s&urce of the label in AQ and IF
DNA has been adequately addressed by our research.
3.1.2 Cross I inked CH^O Located Exclusively in the Interface (IF) DNA
As discussed above, there was £o HPLC evidence for either adducts or
cross-links in the AQ DNA at either 0.3, 2, or 6 ppm. In contrast, the IF DNA
at 6 ppm did provide such evidence.
The Panel states in this section that "the efficiency of extraction of the
DNA from respiratory epithelium under conditions of CrLO dosing (at various
levels) should have been determined1. Detailed information concerning our
extraction efficiencies had not been requested by the Panel prior to the
meeting, nor was it requested subsequent to the meeting. For the record, it
should be noted that the average DNA yield per mg wet weight of respiratory
nucosal tissue (4.20 * 0.10 ftg; mean * SE, n a 30) did not vary significantly
with concentration (£ = 0.473; one-way ANOVA) over an airborne concentration
range of 0.3 to 15 ppm and exposure times of either 3 or 6 hr. Thus, the pos-
sibility of a dose-dependent change in the recovery of DNA is ruled out by our
data.
-6-
-------
Regarding the question of the efficiency of our extraction procedure for
DNA, it should be noted that we used the same extraction procedure to isolate
DMA from rat liver nuclei treated J_n vitro with formaldehyde. The yield of
highly purified ONA (chromatographed on hydroxyapatite and washed by ultrafil-
tration) that we obtained was 0.82 * 0.02 mg/g of liver. This result compares
<»«
extremely well with the total amount of DMA (unpurified) reported to be pre-
sent in rat hepatic nuclei (0.83 * 0.03 mg/g) (Blobel and Potter, 1966). The
(
approximately five-fold higher yield of DMA that we obtained from the nasal
respiratory epithelium (see preceding paragraph) indicates that the percentage
of the total tissue weight that is due to DMA is significantly higher in the
nasal mucosa than in the liver.
3.2 Experimental Methodology Limitations
The Panel again raises questions about the extraction efficiency of the
DMA, and it asserts that the amount of ONA in the IF fraction will vary with
the extraction conditions employed. However, the extraction conditions (vol-
umes, buffers, pH, ionic strength, temperature) were carefully held constant
throughout the experiment. We were well aware of the importance of reprodu-
cibly recovering a constant amount of DNA in all experiments. Consequently,
the same individuals did all of the experiments, and the experiments were
always performed with great care. From the inhalation exposure to the final
extraction of DNA, the experiments were carried out on the same day using the
same protocol. No samples were ever stored. The constancy of the ONA yield
with concentration noted above is clear evidence that our concerns (and those
of the Panel) were properly addressed from the beginning. Therefore, the as-
-7-
-------
section of the Panel of a possibly varying DNA yield with "variations in the
extraction conditions* is contradicted by the evidence.
The question of the interpretation of acute vs. long-term exposure has
been addressed numerous times. The essential point is that at low formalde-
»•
hyde concentrations, _[..<•., those concentrations to which humans are normally
exposed, the transition from normal respiratory to squamous epithelium does
not occur. Therefore, covalent binding studies in normal respiratory epithe-
lial cells are highly relevant to risk assessment, where risk is defined as
the possibility of covalent reaction with DNA (which may or may not lead
eventually to cancer) under normal human exposure situations. The high-con-
centration, long-term exposure studies do not reveal what occurs biologically
at low concentrations, since the cell structure and tissue morphology has been
radically altered. Indeed, one of the panelists, Dr. Wilkinson of Cornell,
remarked at our meeting on December 2 that such high-dose exposures can well
be considered to exceed the "max!mum-to Ierated-dose*.
The final point raised by the Panel in this section is the question of
isotope effects in the oxidation of [3H]- and [14C]CH20. At the time of the
meeting on December 2 we had already begun studies of possible isotope effects
in either the covalent binding or oxidation of CH-0. These studies were star-
ted for reasons other than those raised by the Panel, however, the isotope-
effect studies were only in preliminary stages in December, and, consequent-
ly, the results that we have now obtained could not be given to the Panel. We
would like, therefore, to present these results in this document.
-8-
-------
First, with regard to covalent binding of formaldehyde, we have examined
914
the H/ C ratio of the IF DNA recovered from freshly isolated hepatic nuclei
incubated in vitro with [3H]- and [14C]CH20. The IF DNA isolated from such
nuclei was heavily labeled with H and C, whereas the AQ DNA had practically
no radioactivity, consistent with our interpretation that the IF DNA, and not
the AH DNA, contains covalently bound GO. Furthermore, the percent IF DNA
as well as the specific activity of the IF DNA increased with increasing con-
x^
centra t ions of OLD and with increasing times of reaction. The isotope ratio
of the IF DNA relative to that of the reaction solution was approximately
1.034 * 0.009 (mean * SE, n = 6), indicating an extremely small isotope effect
favoring the binding of [3H]CH20 over that of [UC]CH20 to DNA. An isotope
effect of this small magnitude is negligible insofar as our calculations of
covalently bound GO in DNA are concerned.
•
3 14
Second, with regard to metabolic oxidation, we determined the H/ C ratio
« « *
of [ H]- and [ C]GO at various times during in vitro incubations of selec-
ted concentrations of labeled formaldehyde with freshly isolated homogenates
of the rat respiratory mucosa and NAD (1 mM) . In most cases, the reaction
solutions also contained glutathione (GSH), since the principal enzyme respon-
sible for OO oxidation, formaldehyde dehydrogenase (FDH), is a GSH-requiring
enzyme. We observed that the oxidation of [n]- and [ C]QO catalyzed by
FDH occurs with a significant isotope effect. The rate of oxidation of [ C]-
CHjO was approximately 1.82-fold faster than that of nfJOO, indicating that
the hydride transfer step in the GSH-dependent oxidation of GO to HCOOH
catalyzed by FDH is at least partially rate-limiting in GO oxidation. The
magnitude of the isotope effect was independent of the GO concentration over
-9-
-------
a 100-fold concentration range (from 0.1 to 11 /*M). Furthermore, we found
that aldehyde dehydrogenase exhibits a similar isotope effect to that of FOH
in its oxidation of CHgO to HCOOH, \vhich is GSH-independent.
Since an isotope effect in ChLO oxidation has been demonstrated to occur
in vitro, it can be presumed that it also occurs in vivo. An isotope effect
occurring in the oxidation of OLD to HCOOH in vivo would result in an effec-
tive "enrichment" of [3H]CH20 relative, to [14C]CH20 in the residual (unoxi-
dized) CH20 in the nasal mucosa. The 3H/14C ratio of the unoxidized CHgO
would, therefore, be greater than that of the inhaled gas. In the calculation
of covalently bound CrLO in ONA (Casanova-Schmitz £t aj^., 1984), it was impli-
3 14
citly assumed that the H/ C ratio of QUO in the nasal mucosaI cells was
identical to that of the gas. It is now recognized that this assumption may
be incorrect. The assumption was made because of the practical impossibility
of directly measuring the H/ C ratio of unmetabol ized GO in the cells of
the nasal mucosa.
It can be readily shown that an enrichment of [ HjChLO relative to [ C]-
CH.O in the cells, resulting from an isotope effect in oxidation, leads invar-
iably to an overestimate of the amount of OLD covalently bound to ONA, when
the calculation of covalent binding is done using our published equations
(Casanova-Schmitz et a_L, 1984). Thus, by implicitly assuming ho isotope ef-
fect in CrLO oxidation, our calculation yielded an upper limit on the amount
of OLD bound to DMA. Conversely, by assuming the isotope effect in QLO oxi-
dation to be maximal, j..,e., that it occurs with an isotope effect of 1.82 in-
dependent of the CrLO concentration, we can calculate a Iower limit for co-
-10-
-------
valent binding of GO to DMA. (The reason that 1.82 represents a maximal
isotope effect is that the other routes of CH«0 elimination from cells, such
as covalent binding, diffusion out of the cells, or reduction to rr.ethanol, do
g
not involve breakage of the H-C bond, and, therefore, they would all show
smaller isotope effects than that observed in oxidation. Thus, the assumption
<-
of a maximal isotope effect is equivalent to assuming that metabolism is the
only route of elimination.) The results of such calculations are shown in
Figure 1.
The upper curve in Figure 1 is the same curve as that previously published
by us (Casanova-Schmitz £t jaL, 1984), which assumed no isotope effect in CH«0
oxidation. The lower curve is that which would result if the isotope effect
had its maximal value. Clearly, an overestimate of covalent binding occurs at
all concentrations. However, the fundamental shape of the curves, j..£.» their
significant departure from linearity at low GO concentrations, is unaffected
by the isotope effect in ChLO oxidation.
In reality, by assuming that an isotope effect occurs in vivo in the oxi-
dation of GO, the curve becomes, if anything, even more nonlinear than was
the case before the isotope effect was recognized to occur. This is because
the enrichment of 3H relative to 14C in the residual (unoxidized) OO is
greatest under conditions in which the metabolisa of QO to HCOOH is nost
nearly complete, j.-.e., »t low airborne concentrations, and is smallest under
conditions tn which the metabolism of GO to HCOOH is least complete, i.e.,
at high airborne concentrations (Melander and Saunders, 1980). Therefore, the
overestimate of the amount of GO bound to DMA is greatest at low concentra-
-11-
-------
UPPER AND LOWER BOUNDS FOR FORMALDEHYDE BINDING TO DNA
800-r
[FORMALDEHYDE], ppm
Amounts of CH20 covalently bound to rat nasal mucosal DNA.
Upper and lower curves represent yields of covalently bound CH20
assuming either no isotope effect or a maximal (1.82) isotope
effect in the oxidation of CH20 by FDH.
Pig. 1
-------
tions and is smallest at high concentrations, causing the curve to become even
more nonlinear than before. We would expect, therefore, that the true cova-
lent binding curve should approximate the lower curve of the two shown in
Figure 1 at low GO concentrations, and it should approach more closely to
the upper curve at high CHjO concentrations. We conclude that low-dose non-
^.
linearity in the binding of CH^O to DMA is supported rather than disproved by
our finding of an isotope effect in CH^O oxidation.
3.3 Identity of Labeled Fractions
This issue is addressed in section 3.1 and 3.2.
3.4 Other Measures of Exposure
The Panel remarks that our methods appear to be "unnecessarily sophistica-
ed and complex". We feel that our methods, far from being complex, are rela-
tively simple. DMA was isolated and purified using established (hydroxyapa-
tite) techniques, and the resulting DMA was counted for radioactivity. Cer-
tainly, the use of dual isotopes for metabolic studies is not new. The crit-
icism of our methods by the Panel as being too 'sophisticated" is not justi-
fied in our opinion.
The Panel suggests measuring covalent binding to intraceIlular proteins as'
a "simpler index" of molecular dostmetry. We would ask four questions: (1)
Which intracellular proteins would they recommend? (2) What evidence is there
that covalent binding to proteins is related to mutagenesis or to cancer? (3)
-12-
-------
How would they deal with the issue of protein turnover? (4) How would they
correct for labeling caused by metabolism? Clearly, their proposal requires
selecting a protein, or a group of proteins, that would be used to monitor
CHJO binding. Such proteins would have to be quantitatively purified from the
other proteins 5n the nasal mucosa, and their rate of turnover would have to
<
be separately determined. Labeling due to metabolism would have to be differ-
entiated from that due to covalent binding. Finally, after solving these ex-
tremely challenging experimental problems, it would have to be assumed that
covalent binding to these proteins (presuming that such binding could be shown
to occur) is somehow related to the initiation of cancer. These critical
problems and assumptions clearly mean that intracellular protein binding is
not a "simpler index* than DMA binding for molecular dosimetry purposes.
"x\
Furthermore, we now have strong evidence that protein binding in vivo pri-
marily involves the extracellular proteins. We informed the Panel that when
rats were pretreated with phorone, a GSH depleting agent, there was a marked
increase in the amount of GO covalently bound to DNA, as would be expected
if oxidation of GO to HCOOH were a major defense mechanism (Casanova-Schmitz
and Heck, 1985). In contrast, there was no detectable increase in the amount
of GO covalently bound to proteins as a result of phorone pretreatment.
Therefore, GSH depletion was ineffective in enhancing the binding of GO to
proteins, suggesting that the proteins to. which GO is bound are not protec-*
ted by metabolism. One is led to conclude that the proteins must be either
cell surface proteins or extracellular proteins, the latter of which are, pre-
sumably, mucus proteins. Binding to mucus proteins is irrelevant to dosimet-
ry, since the intracellular concentration of the toxicant, not the extracel-
lular concentration of the toxicant, is of primary concern.
-13-
-------
With regard to the final comments made by the Panel in this Section, we
have never asserted that the amount of CrLO bound to DMA following an acute
exposure is necessarily the same as that following a chronic exposure. We do
not know the amount of CrLO bound under chronic exposure conditions. However,
we do claim that short-term exposure conditions, which do not cause massive
**"
changes in cell structure and morphology, should more nearly represent the
chronic exposure situation at low concentrations of CrLO, j..j».f those to which
humans are actually exposed.
* . *
3.5 Nonlinearity for Cross I inked DMA at Low Doses
3.5.1 Documentation of Nonlinearity for Low Dose Cross I inked DNA
No comment. „_
3.5.2 Alternative Explanations for NonIinear Ity of Low Dose Cross!inked DNA
The Panel suggests that, because each experiment involved only three
replicates, it is possible that the results were spurious. They suggest that
additional experiments below 6 ppm would add considerable substance to the
results. We"are surprised by this comment, since in our meeting with the
Panel on December 2, we did present such evidence to then. As we showed at
that meeting, we have carried out additional studies at 0.9, 2, 4, and 6 ppm
(three replicates at each concentration) (Casanova-Schmitz and Heck, 1985).
The results were fully consistent with those published previously.
-14-
-------
The Panel also suggests the possibility of a 'small but constant loss in
the measurement of IF DNA* as being responsible for low-dose nonlinearity. In
section 3.2, we showed that the Panel's earlier hypothesis of a dose-dependent
loss of DNA was inconsistent with our results. The hypothesis of a dose-
independent loss of DNA from the IF DNA fraction can be readily shown to be
invalid. If one assumes a constant loss of IF ONA at all GO concentrations,
as suggested by the Panel, calculations can be made of the amount of IF DNA
that would have to be lost at 2 and at 6 ppm in order to linearize the cova-
lent binding curve. Such calculations were performed by us: the curve for
covalent binding of CH.O to DNA at 2 and at 6 ppm was recalculated using the
equations in Appendix 3 of Casanova-Schmitz ^t a_L (1984), but allowing for a
constant loss of IF DNA. This loss would affect the binding calculation only
by changing the fraction of DNA that is IF:
Measured (KIF DNA)/100 = (IF DNA)/(AQ DNA + IF DNA);
•True" (JJIF DNAJ/100 • (IF DNA + Loss)/(AQ DNA * IF DNA * Loss).
Figure 2 shows the effect of loss on the ratio of the amount of covalent
binding at 6 ppm to that at 2 ppm. While the ratio is reduced somewhat as the
amount of lost IF DNA increases, in agreement with the suggestion of the Pan-
el, it cannot drop below 8.2 even with arbitrarily large losses. However, in
order to linearize the binding response, this ratio would have to drop to 3,
the ratio of the two airborne formaldehyde concentrations. Therefore, a
constant loss of IF DNA, no matter how large, cannot be responsible for the
observed nonlinearity in ChLO binding to DNA.
-15-
-------
HT
0.0
0.5
1.5
2.0
IF LOSS, MG
Fig. 2. Effect of a hypothetical constant loss of IF ONA on the calculated
ratio of covalently bound ChLO at 6 ppm and at 2 ppm. For linearity of the
concentration-response curve, this ratio should be 3, the ratio of the two
airborne GO concentrations.
-------
3 14
The Panel suggests that "the disproportionate increase in H/ C ratio
with increase in CrLO concentration might be due to artifactual disturbances
...rather than a true increase in crosslinked DMA-protein". What is meant by
such "artifactual disturbances" is not clear, although the Panel appears to be
referring to increases in cell turnover. We have already remarked that in-
*
creases in cell turnover could well contribute to the low-dose nonlinearity in
CrLO binding to DMA (Casanova-Schmitz et al., 1984), just as saturation of
metabolic defense mechanisms could also be a contributing factor. The point
is that a change in the cell turnover rate or a change in CrLO metabolism
would result in a change in the amount of label metabolically incorporated
into DMA. However, the AQ DMA provides a control for such changes, since the
labeling of AQ DMA is only due to metabolism. It is the difference between
the 3H/14C ratios of IF and AQ DNA, not the absolute value of the 3H/14C ratio
of the IF ONA, that is directly related to the amount of covalently bound
CH20.
Thus, the comment of the Panel does not call into question the validity of
our conclusion that the increase in the H/ C ratio of the IF ONA relative to
that of the AQ DNA is due to covalent binding. Rather, it simply suggests a
mechanism for the disproportionate increase in covalent binding at high con-
centrations, J..J9., increased cell turnover, which has already been proposed by
us.
The last comments of the Panel in this section concern the possibility of
cell death, which could (according to the Panel), "inhibit de novo nucleotide
synthesis..., as well as increase the deoxynucleotide pools, thus, diluting
-16-
-------
out radioactivity incorporated into DNA*. The Panel then states that a "de-
3 14
crease in DNA synthesis...would tend to increase the H/ C ratio dispropor-
tionately at higher CH^O concentrations*. We repeat that any change in the
labeling of DNA due to metabolism is already accounted for by our measurement
of the radioactivity in the AQ DNA, which is only due to metabolism and is,
<»
therefore, an internal control for all such "effects. Thus, the Panel has
merely suggested a possible mechanism for low-dose nonlinearity. We doubt
that this mechanism is plausible, however, because at low CIO concentrations
(0.3, 2, and 6 ppm) where the nonlinearity occurs, cell death is not an im-
portant consideration. We know from the work of Dr. Kevin Morgan at CUT that
nasal mucociliary activity continues even after many days of exposure to 6 ppm
of CH20.
3.6 Sensitivity of the Study Conclusions to Statistical Analysts
This section deals only with our statistical analyses of labeling, j..£.,
the C-specific activity and the normalized H/ C isotope ratios in the IF
and AQ DNA. The two-way analyses of variance of these measurements reported
in Casanova-Schmitz et aJK (1984) did not account for pairing between the IF
and AQ DNA determinations or possible inhomogeneity of variance as a function
of the exposure concentration. Furthermore, we did not specify ordered (over
concentration) alternatives to the null hypothesis of no treatment effects
prior to our examination of the data. Consequently, as noted by the Panel,
these analyses had somewhat less than optimal power to detect systematic dif-
ferences in treatment effects. Despite this limitation, statistically signi-
ficant effects of exposure concentration and DNA fraction (including interac-
-17-
-------
tion between these two factors) were detected in both the C specific activ-
3 14
ity and the normalized H/ C isotope ratios. As has been verified subse-
quently, more sensitive statistical analysis procedures do no more than con-
firm these findings. Thus, as was noted by the Panel, our overall conclusions
regarding these measurements are valid.
In the second and third paragraphs of this section, the C specific ac-
tivity measurements are described by the Panel as 'bound ( C) CH-O*. This
confusing terminology can easily give readers the mistaken impression that
covalent binding to DNA, rather than total C specific activity in the two
DNA fractions, is being discussed. We therefore strongly suggest changing
this phrase to ' C specific activity" or 'total C specific activity".
The absence of a statistically significant difference between the C
specific activities of the IF and AQ DNA fractions at 0.3 and 2 ppm is des-
cribed by the Panel as "limiting the extent to which inferences can be made
about the responses to low concentrations for purposes of identifying no-
response levels or making low-dose extrapolations for risk assessment." We
disagree with this statement. Such extrapolations should not be based solely
on the difference between the C specific activities of IF and AQ ONA, when
it is the difference between the 3H/14C ratios (not the 14C specific activity
difference) that is the-primary indicator of covalent binding to DNA. Signi-
3 14
ficant differences between the H/ C ratios of the two DNA fractions were
observed at all concentrations equal to or greater than 2 ppm. Consequently,
the concentration of CH-O covalentIy bound to DNA was found to differ signi-
ficantly from zero at these concentrations as well.
-18-
-------
3.7 Adequacy of the Measure of Exposure
No comment.
3.8 Utility of the Study in the Quantitative Risk Assessment of CH00
_. .
The Panel concludes that our results should not be used for risk assess-
ment. Their conclusion is based on their arguments that: (1) experimental
methodologies have not been properly 'validated"; (2) intracellular proteins
rather than ONA should be used as the target; and (3) acute exposures may not
be relevant to chronic exposures. We disagree with the Panel on all points.
First, the experimental methodologies have been validated, both by HPLC
analysis and by repetition of the experiments, as discussed above in Sections
3.1.1 and 3.5.2. We demonstrated the lack of variation of the DNA yield with
concentration in Section 3.1.2, and we have determined the magnitude of the
isotope effect in CH^O oxidation and discussed its implications in Section 3.2
(the binding curve becomes more nonlinear rather than less). We showed in
Section 3.5.2 that the assumption of a constant loss of DNA from the IF DNA
fraction cannot account for low-dose nonlinearity no matter how large the loss
is assumed to be. Finally, we discussed the fact that all experiments were
carefully controlled to maintain constant extraction conditions for DNA at all
exposure concentrations in Section 3.2.
Second, the argument that intracellular proteins should be used for dosi-
metry fails to recognize: (1) that covalent binding to proteins is not widely
-19-
-------
acepted as a mechanism of mutagenesis, (2) that proteins have a much higher
rate of turnover than DMA, and that individual proteins vary in their turnover
rate, (3) that correcting for labeling of the proteins due to metabolism pre-
sents major experimental problems, and (4) that, in any case, protein binding
occurs primarily on the extracellular proteins, not on the intracellular pro-
teins. These points are discussed in detail in*Section 3.4.
Third, the rationale for using acute exposure data for risk assessment at
low concentrations is presented in section 3.2. These exposure conditions do
not cause massive toxicity to the tissue, and, therefore, more closely repre-
sent the actual human chronic exposure situation. The results obtained pro-
vide information about the ability of CH-O to react with DMA under realistic,
low-1 eveI exposure conditions.
In summary, we find the concerns of the Panel to be without merit and
their conclusions to be unsubstantiated.
-20-
-------
REFERENCES
Beland, F. A., Fuller-ton, N. F., and Heflich, R. H. (1984). Rapid isolation,
hydrolysis, and chromatography of formaldehyde-modified ONA. J. Chromatogr.
308, 121-131.
*•
Blobel, G., and Potter, V. R. (1966). Nuclei from rat liver: isolation method
that combines purity with high yield. Science 154, 1662-1665.
Casanova-Schmitz, M., and Heck, H. d'A. (1985). DNA-protein cross-linking in-
duced by formaldehyde (FA) in the rat respiratory mucosa: dependence on FA
concentration in normal rats and in rats depleted of glutathione (GSH). The
Toxicolegist S. 128 (Abst. |509).
Casanova-Schmitz, M., Starr, T. B., and Heck, H. d'A. (1984).-Differentiation
between metabolic incorporation and covalent binding in the labeling of macro-
molecules in the rat nasal mucosa and bone marrow by inhaled [ C]- and [n]-
formaldehyde. Tox i eoI. AppI. Pharmacol. 76, 26-44.
Gehrke, C. W., Kuo, K. C., McCune, R. A., and Gerhardt, K. 0. (1982). Quanti-
tative enzymatic hydrolysis of tRNAs. Reversed-phase high-performance liquid
chromatography of tRNA nucleosides. J. Chromatogr. 230. 297-308.
Kornberg, A. (1980). DMA Rep Iication. Freeman, San Francisco, pp. 644-646.
Melander, L., and Saunders, W. H., Jr. (1980). Reaction Rates of Isotopic Mol-
ecules. Wiley, New .York, p. 99.
-21-
-------
Shapiro, H. S. (1968). Distribution of purines and pyrimidines in deoxyribo-
noclcic acids, in Handbook of Bioch«mS.f.rr ffi A. Sober, ed.), Chemical Rubber
Co., Cleveland, Ohio, pp. H:30-H:51.
-22-
-------
APPENDIX 2: INDIVIDUAL SUMMARIES OF EPIDEMIOLOGIC
STUDIES REVIEWED
-------
Matanoski (1982) of Johns Hopkins University examined
mortality patterns of male pathologists in 2
professional societies, American Associations of
Pathologists and Bacteriologists (AAPB) and the American
Society for Experimental Pathology (ASEP). The results
of these analyses were reported in a March 30, 1982
letter to John Martonik, Deputy Director of the
Occupational Safety and Health Administration (OSHA).
In separate analyses of each group, Matanoski compared
the causes of death to the number of expected deaths
using U.S. white males and using psychiatrists as the
referent group. Additionally, Matanoski combined the
two pathologist groups without overlap and compared
proportions of deaths to those expected proportions
using 1) psychiatrists and 2) internists,
otolaryngologists, and opthalmologists as the referent.
In the SMR analysis, Matanoski observed the same
pattern of deaths when either referent group was used as
the standard. The present review has focused on results
from comparisons with psychiatrists. For ASEP members,
apparent excesses were observed for neoplasms of the
liver (SMR=399, 3 observed), pancreas (SMR=277,
5 observed), and lymphomas (SMR=272, 3 observed).
Deficits were observed for all deaths, all neoplastic
deaths, and neoplasms of the lung, kidney, and bladder,
brain, and lymphopoietic system. None were
significant. For AAPB members, Matanoski observed
-------
elevated mortality (not statiscally significant) from
esophageal and small intestinal neoplasms (SMR=156,
2 observed), pancreas (SMR=263, 9 observed), brain
(SMR=296, 5 observed), and lymphoma/multiple myeloma
(SMR=174, 3 observed). Deficits were seen for all
causes of deaths, all cancer deaths, and neoplasms of
the stomach, large intestine, prostate, and
lymphopoietic system. These deficits were not
statistically significant.
Findings from the PMR analysis support the above
observations. Matanoski, in addition, observed a
statistically significant increase in the proportions of
deaths due to neoplasms of the hypopharynx (PMR = 3060,
p_<_0.005, 2 observed).
Harrington and Shannon (1975) of the London School of
Hygiene and Tropical Medicine conducted a SMR analysis
of 2,079 pathologists who were members of the Royal
College of Pathologists or the Pathological Society of.
Great Britain during the period of 1955 to 1973. By the
end of 1973, 156 deaths occurred. The authors reported
a significant excess in mortality from lymphopoietic
system cancers (SMR=200, 8 observed, p<0.05),
particularly from lymphatic and hematopoietic diseases •
not due to Hodgkin's disease or leukemia (SMR=353,
6 observed, p<0.01). Additionally, increased mortality
from Hodgkin's disease appeared (SMR=167, 1 observed)
for male pathologists in England and Wales.
- 2 -
-------
Harrington and Shannon also presented an analysis
of 12,944 laboratory technicians who were registered
with the Council for Professions Supplementary to
Medicine. Between August, 1963 and December 31, 1973,
154 deaths occurred. Deficits of deaths were observed
from all causes and all neoplastic causes, including
neoplasms of the digestive tract and peritoneum, lung,
lymphohematopoietic system, Hodgkin's disease, and
leukemia. These observations were not statistically
significant. The most striking statistically
significant excess in mortality was observed from
suicides (SMR=243, 17 observed, p<0.001). In addition,
Harrington and Shannon observed some excess in
lymphohematopoietic deaths (not including leukemia or
Hodgkin's disease) (SMR=118, 2 observed). This
observation was not statistically significant.
Harrington and Oakes (1982) more recently followed the
Royal College of Pathologists' cohort from 1974 to 1989
and performed an SMR analysis of 2,720 members (2,307
males and 413 females) in which 126 total deaths (110
males, 16 females) occurred. Harrington and Oakes
observed increased mortality in males from cancers of
the brain (SMR=331, 4 observed, p<0.05) and bladder
(SMR=107, 2 observed), from accidents (SMR=170,
13 observed p<0.05), and from suicides (SMR=353,
7 observed p<0.01). Increased mortality from lymphatic
and hematopoietic neoplasms was not reported for male
- 3 -
-------
pathologists (2 observed deaths) but was reported for
female pathologists (SMR=370, 1 observed death).
Deficits in mortality (statistically significant,
p<0.05) were observed for males from all neoplasms
(SMR=61, 32 observed) and from neoplasms of the lung
(SMR=41), 9 observed) and digestive and peritoneal
system (SMR=51, 8 observed). All the malignant brain
tumors diagnosed in males were of the astrocytoma/glioma
cell type.
4. Levine et al. (1984) of CUT in an SMR analysis found
excess mortality among Ontario morticians, relative to
Ontario white males, from lymphopoietic cancer (SMR=124,
8 observed), particularly, leukemias/aleukemias
(SMR=160, 4 observed), and brain cancers (SMR=115, 3
observed). None of these malignancies was significantly
elevated. Only cirrhosis of the liver and rheumatic
heart disease (SMR=199, 8 observed, p<0.05) showed
significant excesses (SMR=171, 18 observed, p<0.001).
Decreases in mortality (not statistically significant)
were observed for neoplasms of the lung, digestive
system, and buccal cavity.
In an earlier analysis of these deaths where U.S.
white males were used as the standard population, Levine
observed increasing SMR's with increasing time since
first exposure for cancers of the brain, lymphopoietic
system, and leukemia/aleukemia.
- 4 -
-------
Levine et al. did not report exposure levels to
which Ontario undertakers may be exposed, but data from
a survey of seven West Virginia funeral homes were
presented. Mean time-weighted averages for breathing
zone samplers showed HCHO levels between 0.3 ppm and
0.9 ppm (Williams et al., 1984). It is not known how
similiar the exposures are between these 2 groups.
5. Stroup (1984) noted excesses, when compared to U.S.
.white males, in mortality due to brain cancers (SMR=271,
10 observed, p<0.01) and leukemias (SMR=148, 10
observed) in her unpublished SMR study of 2,239
anatomists who were members of the American Association
of Anatomists (AAA). Stroup noted excesses of the cell
types astrocytoma/glioblastoma (all ten brain tumors)
and myeloid leukemia (5 of the 10 leukemia deaths).
Deficits in mortality were observed for neoplasms of the
lung (SMR=28, 12 observed, p<0.05), buccal cavity and
pharynx (SMR - 15, 1 observed, p<0.05) and of the nasal
cavity or sinuses (0 observed, 0.14 expected).
Stroup further examined the relationship between
exposure and mortality from brain cancer and leukemia.
The number of years of membership in the AAA was used as
a surrogate for exposure. In these analyses, only the
SMR's for brain cancer increased as the membership years
increased. To further examine the relationship between
mortality from the above three neoplasms and exposure to
formaldehyde, Stroup categorized the anatomists'
- 5 -
-------
subspecialities by their potential usage of HCHO. Gross
anatomists were classified as having high HCHO exposure,
anatomists who had specialities in both gross anatomy
and microanatomy had medium exposure, and
microanatomists had low HCHO exposure. No trend with
formaldehyde rank was observed. This analysis, however,
may be limited by the small numbers of deaths (less than
5 in each category) and may be biased due to exposure
misclassification.
To examine if social class differences or if
ascertainment may have biased the observed excess brain
cancer and leukemia mortality, Stroup used psychiatrists
as a comparison group. In this comparison, the excess
brain cancer and leukemia mortality remained (brain
cancer, SMR=572, p<0.01; leukemia, SMR=212,
p<0.05). It can be concluded from these analyses that
neither social class differences nor ascertainment bias
accounts for the observed increased in brain cancer and
leukemia mortality.
6. Wong (1983) observed that among 2,026 workers employed
by Celanese in a HCHO manufacturing plant, as compared
to U.S. males, mortality was increased (not
statistically significant) from neoplasms of the skin
(SMR=109, 95% CI*:2-717, 1 observed), bone (SMR=430, 95%
CI:6-2751, 1 observed), prostate (SMR=305, 95% CI:84-
797, 4 observed), bladder (SMR=122, 95% CI:2-705,
*
* 95% Confidence Interval (95% CI)
- 6 - •
-------
1 observed), kidney (SMR=102, 95% CI:1-613, 1 observed),
brain (SMR=186, 95% CI:43-623, 3 observed), and
lymphopoietic system (SMR=136, 95% CI:43-623, 3 ob-
served), and lymphopoietic system (SMR=136, 95% CI:57-
338, 6 observed), including Hodgkin's disease (SMR=240,
95% CI:33-1063, 2 observed) and leukemia/aleukemia
(SMR=118, 95% CI:15-487, 2 observed). Lung cancer
mortality was decreased (SMR=82, 95% CI:37-156, 9 ob-
served) and no nasal cavity or sinus neoplastic deaths
were observed. Accounting for a latency of 20 years,
Wong observed significantly increased mortality from
cancer of the prostate (SMR=431, 4 observed, p<0.05) and
apparently increased mortality from lymphopoietic system
(SMRS=231, 95% 01:62-591, 4 observed), including
Hodgkin's disease (SMR=582, 95% CI:8-3236, 1 observed).
Again, lung cancer mortality was decreased (SMR=87, 95%
CI:32-190, 6 observed). Besides HCHO, this cohort had
potential exposures to other oxygenated hydrocarbons,
benzene, asbestos, and inorganic and organic pigments.
Exposure to benzene is particularly important since the
literature reports a causal association between leukemia
and benzene exposures (Heath, 1982).
Tabershaw Associates (1982) studied the same cohort as
Wong, with 58 men added who had incorrectly been
excluded and with the HCHO-exposed workers identified.
An SMR analysis of the exposed and unexposed cohorts and
a case-control analysis using randomly-selected controls
- 7 -
-------
among the non-cancer cases were conducted. In the SMR
analysis of 867 HCHO-exposed workers, increased
mortality (not statistically significant) from prostatic
(SMR=364, 2 observed), brain/CNS (SMR=135, 1 observed)
cancers and from all accidents (SMR=103, 11 observed)
was reported. Interestingly, Tabershaw Associates base
the brain/CNS conclusion on one observed death, yet the
text indicates that 2 men who had 6.7 years and 24 years
of exposure died of this cause. Decreased mortality
from lung cancer (SMR=54, 3 observed) was observed along
with no deaths from neoplasms of the digestive organs
and peritoneum, nasal cavity and sinuses, and bladder.
In the case-control analysis, increased odds ratios
for cancers of the prostate (OR=2.67) and lymphopoietic
system (OR=3.0), and for all neoplasms (OR=1.2) were
reported for cases with 5 to 15 years of HCHO
exposure. These increases were not statistically
significant. Risks did not appear to increase with
increasing years of exposure or increasing number of
years employed at the plant. Note that Tabershaw
Associates did not use an unexposed group as a
comparison, but compared the exposed employees to those
with less than 5 years of exposure. The use of this
group as the "controls" may diminish the ability of this
analyses to detect a small elevation in risk due to a
higher background prevalance of formaldehyde exposure.
- 8 -
-------
8. Acheson (1984a) of MRC Environmental Epidemiology Unit,
Southampton General Hospital, in an ongoing study of
7,716 workers in six plants which use or manufacture
HCHO, has observed significant decreases in overall
mortality (SMR=87, 1,619 observed, p<0.05) and
nonsignificant increases from buccal cavity and
pharyngeal (SMR=121, 5 observed), esophageal (SMR=103,
13 observed), respiratory (SMR=102, 236 observed), and
lung (SMR=105, 205 observed) cancers.
Acheson et al. subjectively categorized exposure on
the basis of workers' recall of acute irritation. These
categories were defined as: nil/background, <0.1 ppm;
low, 0.1-0.5 ppm; moderate, 0.6-2.0 ppm; high, >2.0 ppm.
In analyses based on these categories, Acheson et al.
found a significant excess of bone cancer mortality and
a significant dose-response relationship for .lung cancer
mortality in one plant (BIP), the cohort with the
highest exposure. In a comparison with local controls,
the dose-response relationship (marginally significant)
was still observed. Acheson lacked smoking histories
for the entire cohort, and the BIP plant is located in
the West Midlands area, an industrially polluted area
with high referent lung cancer rates. The use of a
local comparison may have overestimated the number of
expected lung cancer deaths (Enterline, 1976) since the
local lung cancer rates may be influenced by the BIP
lung cancer deaths; reducing the power of the analysis.
- 9 -
-------
In subsequent analysis of the mortality data for
lung cancer among individuals employed at this plant,
Acheson et al. (1984b) observed that the risks for lung
cancer did not increase with duration of employment or
with cumulative doses (as assessed by three measures).
In a third analysis of all 120 lung cancer deaths
and 640 controls in the BIP plant, Acheson et al.
(1984c) examined smoking history and previous
employment. Acheson et. al. found no differences between
the cases and their controls. Only 11% of the cases and
12% of the controls had adequate information on smoking,
however.
9. Marsh (1983a) of the University of Pittsburgh conducted
an SMR analysis and a case-control study nested within
the cohort study of Monsanto chemical workers. This
plant produced plastics and workers had potential
exposures to HCHO, vinyl chloride, styrene, and
cellulose acetate. Marsh compared the mortality
experience of all workers to the white male populations
of the U.S., of Massachusetts, and of Hampden County,
the county from which the workforce was drawn. In the
SMR study, the cohort consisted of 2,490 male workers
with a minimum of one year employment. Among the 2,490
workers, 591 deaths were identified by the company or by
death certificate searches. Marsh reported increased
mortality (not statistically significant) due to all
neoplasms (SMR=107, 127 observed). Among all neoplasms,
- 10 -
-------
excess mortality was observed from cancer of the buccal
cavity and pharynx (SMR=155, 6 observed), digestive
organs and peritoneum (SMR=126, 44 observed), prostate
(SMR=178, 14 observed), bladder (SMR=135, 5 observed),
genitourinary tract (SMR=169, 26 observed, p<0.05),
Hodgkin's disease (SMR=118, 2 observed), and all other
lymphopoietic tissue (SMR=153, 4 observed). No
relationship was observed between genitourinary system
neoplasms and length of employment.
In the matched case-control study based on the
cancer deaths, Marsh presented odds ratios for digestive
system, rectal, genitourinary, and prostatic cancers and
21 occupational exposure categories. Two of the
21 categories had pertinent exposure to HCHO either as a
chemical (resin production) or in a product (resins
processing). Marsh observed increased odds ratios for
digestive system cancer in the resins processing
category (OR=1.83) and for rectal cancer in both
categories (resins production, OR=3.75; resins
processing, OR=2.00). These increases were not
statistically significant. All cases in the
occupational categories had from 1 month to 5 years
exposure and increasing risk was not observed with
increasing duration of exposure.
10. Bertazzi et al. (1984) of the Institute of Occupational
Health, University of Milan, presented at the 3rd
International Conference on Epidemiology and
- 11 -
-------
Occupational Health findings of a cohort study of HCHO
resin manufacturing workers. The mortality experience
of 1,332 male employees who had worked six (6) months or
more between 1959 and 1980 was compared to the expected
number of deaths using national and local rates.
Bertazzi et al. noted that ambient monitoring showed
many work areas were above the Threshold Limit Value.
Area sampling values between 1974 and 1979 ranged
between 0.2 and 3.8 mg/m^.
For the entire cohort, Bertazzi et al. observed
significantly increased mortality for all neoplasms
(SMR = 154, 42 observed, p<0.05 national rates) and for
lung cancer when both national (SMR = 236, p<0.05 18
observed) and local (SMR = 186, p < 0.05) rates were
used as the referent. Mortality from digestive (SMR =
156, 14 observed, national rates) and lymphopoietic (SMR
= 201, 5 observed, national rates) and esophageal-
stomach (SMR = 148, 70 observed, national rates)
neoplasms were apparently elevated.
Bertazzi et al. had work histories for all but
18 percent of the cohort. Using local rates as the
standard, Bertazzi et al. examined mortality among
formaldehyde-exposed workers. In this analysis,
mortality appeared elevated from all causes (SMR = 111,
51 observed), and all neoplasms (SMR = 128, 19
observed), particularly alimentary tract (SMR = 155,
8 observed), lung (SMR = 136, 5 observed) and
- 12 -
-------
hematologic (SMR = 273, 3 observed) neoplasms. In
analyses examining cause-specific death by duration of
exposure, only the SMR's for lung cancer increased as
the number of years employed increased. Statistical
testing was not performed to see if this trend was
significant.
11. Blair et al. (1986) of the National Cancer Institute and
the Formaldehyde Institute conducted a SMR study of
26,561 workers in 10 plants which manufacture or use
formaldehyde; 7 plants produced resins, 2 plants
photographic films, and one plant produced plywood. Two
of the 10 plants were included in the studies of Wong
(1983), Tabershaw Associates (1982), Marsh (1983a and
1983b), and Liebling et al. (1984). The Blair et al.
study cohort was the largest ever studied for
formaldehyde exposure. Any worker who had ever been
employed in any one of the 10 plants was included in the
cohort. Over 80 percent of the cohort were white males.
Blair et al. examined the relationship between
mortality and exposure as categorized 3 ways:
1) time-weighted average (TWA), 2) peak exposure, and
3) cumulative exposure in ppm-years.
In the analysis of TWA exposure, Blair et al.
examined the mortality experiences of white males with
TWA exposures of >0.1 ppm (exposed) and compared their
experience to the mortality experience of white males
with TWA exposure of jc_0.1 ppm (nonexposed) . Elevations
- 13 -
-------
in the SMR over 100 for the exposed appeared present for
cancer of the prostate (SMR=115, 33 observed), liver
(SMR=102, 11 observed), lung (SMR=111, 201 observed),
bone (SMR=123, 40 observed) and kidney (SMR=123,
18 observed) and for Hodgkin's disease"(SMR=142, 14
observed, while the SMR's of the nonexposed were less
than 100. None of the elevated SMR's was statistically
significant. In the analysis of peak exposure, only
Hodgkin's disease among white males showed a significant
increasing trend with intensity of exposure. This trend
also appeared present, but was not statistically
significant, when cumulative exposure was examined.
In analyses examining cumulative exposure, Blair et
al. observed among white males a significant elevation
of neoplasms of the lung (SMR=122, 212 observed,
p_<_0.05) and nasopharynx (SMR=300, 6 observed, p_^0.05)
whose cumulative exposure to formaldehyde was greater
than 0 ppm-years. A trend with increasing exposure was
not reported for either site. When latency, defined as
>20 years, was accounted for, the lung cancer excess
increased and remained significant (SMR=135, 146 ob-
served, p<0.05), but the nasopharyngeal excess became
marginally significant (SMR=300, 3 observed, p = 0.08).
Again, no trend with increasing exposure was observed
for either site.
Exposure characterization for early exposures may
be subject to recall and misclassification bias,
- 14 -
-------
although an elaborate exposure matrix was developed for
this study. Historic exposures were estimated from
sensory perception, previous monitoring, and current
levels with knowledge of plant process changes. Current
levels were determined using three methods, NIOSH P&CAM
125 area monitoring, and DuPont and 3M passive
dosimeters. The use of passive monitors for low level,
short term exposures may not be valid and their use,
even weighted, may not be appropriate. Second,
formaldehyde exposure levels for several jobs were below
the analytical method's level of detection, with the job
identified as having an exposure of 0.0 ppm. Thus,
weighting the exposures associated with 0.0 ppm (i.e.,
below the analytical method's level of detection) may be
inappropriate. Third, industrial hygiene data show wide
variation in formaldehyde levels for a single given
job. Fourth, sensory perception to formaldehyde may be
influenced by recall. All of these factors compound to
reduce the certainty of the exposure categorization.
Blair et al. argue that this study provides little
support that formaldehyde exposure is associated with
cancer. The authors based this conclusion on the lack
of a dose-response relationship between exposure and
lung and nasopharyngeal neoplastic mortality. Apparent
lack of an exposure-dose trend cannot diminish the
importance of the 30% excess in lung cancer mortality
and the 200% excess in nasopharyngeal mortality, for
- 15 -
-------
these are statistically significant increases. Smoking
may not account for the observed excesses in lung cancer
mortality; when the SMR's across exposure groups for
lung cancer are compared with another smoking-related
endpoint, emphysema, the SMR's for emphysema decrease
with increasing exposure whereas the SMR's for lung
cancer remain elevated.
There may be several reasons for the observed lack
of a dose-response trend in this study. Most
importantly, misclassification and recall bias may be
present. Second, even though workers who began
employment in the 1930's and 1940's are included in this
cohort, 44% of the cohort entered the study between 1956
and 1965. Since vital status was obtained in 1980, a
full latency period for these workers may not have been
obtained and may bias the results towards the null
hypothesis of no effect.
Subsequent to the release of this study, OTS/EPA
has received notice that 4 of the 6 nasopharyngeal
cancer deaths occurred at 1 plant which manufactured
resins and molded compounds. There is a. significantly
elevated SMR for this plant (SMR=920, 4 observed,
p<0.01). All 4 workers died 15 or more years since
first exposure and all 4 deaths had worked in the early
part of their employment in the same position. The only
other pertinent exposure besides formaldehyde was to
cellulose pulp dust (personal conversation with Dr. Jim.
- 16 -
-------
Collins, Manager of Epidemiology, American Cyanamid
Company).
Blair et al. (1987) performed analyses of the
nasopharyngeal and oropharyngeal cancer deaths which
examined particulate exposure. In these analyses, Blair
et al. observed for those workers with particulate
exposure an apparent increasing trend between
nasopharyngeal cancer mortality and cumulative
formaldehyde exposure, however, this trend was not
statistically significant at a p=0.05 level. No trend
was seen between nasopharyngeal cancer mortality and
cumulative formaldehyde for those workers not exposed to
particulates nor was there a trend for oropharyngeal
cancer by cumulative formaldehyde, regardless of
particulate-exposure status. These analyses are limited
by the small number of deaths in the subcategories. In
addition, it is possible that the nasopharyngeal cancer-
formaldehyde dose-gradient, is a surrogate for an
unmeasured particulate gradient. Blair et al. (1987)
believe, however, that formaldehyde and particulates
together appear to be a risk factor for nasopharyngeal
cancer. Blair et al. (1987) postulated that the
delivered formaldehyde dose for these workers may be
higher than was estimated in Blair et al. (1986) due to
formaldehyde attachment to the particulate matter.
12. Stayner et al. (1986) of the National Institute of
Occupational Safety and Health (NIOSH) conducted a
- 17 -
-------
cohort mortality study of 11,030 workers in 3 garment
facilities that used formaldehyde resins in the
production of permanent press garments. Two of the
three facilities were included in the proportionate
mortality analyses of Stayner et al. (1985). In this
analysis, which is described in #17, only the
proportions of deaths in 3 plants were analyzed. In the
present cohort study, 1 plant in the PMR was replaced
with another larger plant, in terms of the number of
employees, from another company.
Garment workers included in the study cohort must
have worked for at least 3 months between the time when
formaldehyde fabrics were first introduced into the
production process and December 31, 1977. The cohort
was composed mainly of workers who were white women,
were from plant 1, had been employed from 3 months to
4 years, and had first exposures before 1963.
Free formaldehyde was extensively measured by NIOSH
investigators between 1981 and 1984 using both area
samplers and personal monitors. The time-weighted-
average HCHO level for these years was 0.15 ppm (a range
of 0.14-0.17). Stayner et al. sampled for potential
confounding exposures such as phenol, organic solvents,
and dust. These industrial hygiene surveys did not
identify any chemical exposures which could result in
substantial confounding. Likewise, nuisance dust levels
were minimal.
- 18 -
-------
A total of 609 deaths (822 expected; SMR=74) were
observed among 188,025 person-years. No significant
deficits in site-specific deaths were noted by the
authors. Stayner et al. observed statistically elevated
excesses in deaths due to cancer of the- buccal cavity
(4 observed, SMR=343, p<0.05), tonsils (2 observed,
SMR=694, p<0.05), and connective tissue (4 observed,
SMR=364, p<0.05). The statistical test employed by
Stayner et al. is one-sided since the investigators had
a_ priori planned to examine the relationship between
formaldehyde exposure and elevations in cancer mortality
(not whether formaldehyde exposure was related to change
in cancer mortality, use of a two-sided test
statistic). All 4 of the buccal cavity deaths were
among females whose first exposure was between 1955 and
1962. Two of the 4 buccal cavity cancer deaths were
parotid tumors, and the other 2 deaths were cancers of
the oral mucosa and soft palate.
There appeared to be an increase in mortality from
neoplasms of the lung, trachea, and bronchus
(39 observed, SMR=114) and other lymphopoietic sites
(5 observed, SMR=170), from leukemia (9 observed,
SMR=114), and from bronchitis (SMR=190, n=4). Mortality
from neoplasms of the brain appeared decreased (5 ob-
served, SMR=71). No deaths from nasal cancer were
observed in this cohort.
- 19 -
-------
Stayner et al. analyzed the data by plant, by
length of latency, and by duration of exposure. In
these analyses, Stayner et al. observed elevated
mortality from cancers of the trachea, bronchus, and
lung (29 observed, SMR=149, p<0.05) and connective
tissue (3 observed, SMR=514, p<0.05) in plant 1 only.
Stayner et al. noted that the remaining 2 plants each
lacked sufficient power to detect small to moderate
elevations in lung cancer risks. In the analyses which
examined duration of exposure and latency period,
Stayner et al. observed the highest excesses in
mortality from cancers of the buccal cavity, connective
tissue, and other lymphopoietic tissue among those
workers with the longest duration of exposure (10+
years) and with the greatest latency period (20+
years). In these analyses, the risk of lung cancer
appeared to decrease with increasing duration of
exposure. An EPA analysis of the data in the paper
shows there is a statistically significant increasing
trend (p < 0.05) between buccal cavity cancer mortality
and duration of exposure, although such an analysis had
not been performed by Stayner et al.
Stayner et al. concluded that the excesses in
mortality from buccal cavity neoplasms, leukemia, and
other lymphopoietic neoplasms were consistant with the
hypothesis of being formaldehyde related. It is not
known how the excess in connective tissue cancer
- 20 -
-------
mortality may relate to exposure. It must be recognized
that these findings are based on a small number of cases
and that confounding with other exposures may exist.
The investigators believe, however, indirect evidence
suggests that cigarette and alcohol consumption were not
confounders for the observed excess in buccal cavity
cancer mortality.
13. Walrath and Fraumeni (1983) of NCI conducted a PMR study
of 1,132 funeral directors or embalmers licensed in New
York. In this cohort, Walrath and Fraumeni observed
significantly elevated mortality from skin (PMR=221,
8 observed, p<0.05) and colon (PMR=143, 29 observed,
p<0.05) neoplasms. Elevations also appeared present for
cancer of the buccal cavity and pharynx (PMR=113,
8 observed), digestive system (PMR=104, 68 observed),
liver (PMR=106, 5 observed), pancreas (SMR=105, 13 ob-
served), lung (PMR=108, 72 observed), brain/CNS
(PMR=156, 9 observed), kidney (PMR=150, 8 observed), and
lymphatic/hematopoietic system (PMR=121, 25 observed).
No nasal cavity and sinus neoplasms were observed.
Among those licensed as embalmers only, Walrath and
Fraumeni observed increases in mortality from buccal
cavity and pharyngeal (PMR=201, 7 observed), skin
(PMR=326, 5 observed, p<0.05), and brain/CNS (PMR=234,
6 observed, p<0.05) cancers. In the analysis for
latency, Walrath and Fraumeni observed, for the entire
- 21 -
-------
cohort, increasing PMR's for skin (p<0.05) and brain/CNS
neoplasms with increasing time since first licensed.
Walrath and Fraumeni did not report actual exposure
data for these embalmers, but data from previous
industrial hygiene surveys were presented. In a NIOSH
survey of a mortuary science college, formaldehyde
levels ranged between 0.2 and 0.9 ppm. Another survey
of 6 funeral homes reported formaldehyde levels ranging
from 0.1 to 5.3 ppm.
14. Walrath and Fraumemi (1984) conducted another PMR
analysis of 1,050 embalmers in California and observed
similar findings as from the analyses of N.Y.
embalmers. Walrath and Fraumeni reported significantly
increased mortality from neoplasms of the brain
(PMR=193, 9 observed, p<0.05), leukemia (PMR=175, 12
observed, p<0.05), and prostate (PMR=176, 23 observed,
p<0.05). Increases that were not statistically
significant were also reported for lymphatic and
hematopoietic system (PMR=123, 19 observed) and buccal
cavity and pharyngeal (PMR=131, 8' observed) cancers.
Mortality from lung neoplasms was slightly decreased
(SMR=96, 41 observed). As in the previous Walrath and
Fraumeni study, no neoplasms of the nasal cavity and
sinuses were reported.
Walrath and Fraumeni did not report actual exposure
data for these embalmers, but the above-mentioned
industrial hygiene data were included.
- 22 -
-------
15. Marsh (1983b) of the University of Pittsburgh conducted
a PMR analysis of HCHO-exposed workers at the Monsanto
plant described previously. Marsh found 136 deaths
among male workers with exposure of 1 month or greater
in a "formaldehyde related plant area". Marsh compared
their mortality experience to U.S. male, age-race
adjusted, proportionality mortality data.
In the HCHO-exposed white males (115 deaths), Marsh
observed increased (not statistically significant)
mortality from cancers of the genitourinary system
(PMR = 121, 3 observed), including the bladder (PMR =
330, 2 observed) and of the digestive organs and
peritoneum (PMR = 127, 8 observed), particularly the
pancreas (PMR = 160, 2 observed). Marsh observed
decreased mortality from all neoplasms (PMR = 90,
20 observed), particularly the respiratory system
(PMR = 80, 2 observed).
Only 21 deaths occurred among non-white
formaldehyde-exposed males, and 2 of these deaths were
from neoplasms (1 respiratory, 1 genitourinary). Marsh
does not report PMR's for those sites where less than 2
deaths occurred.
In this study, Marsh additionally examined cause-
specific mortality for those workers not exposed to
formaldehyde. The most striking observations were of
significant excesses in deaths from genitourinary tract
neoplasms (white males), (PMR = 192.3, 22 observed,
- 23 -
-------
p<0.05) and from all neoplasms (non-white males),
(PMR = 251, 5 observed, p<0.01), particularly other
malignant neoplasms (PMR = 882, 3 observed, p<0.01).
Marsh does not present exposure information for the
formaldehyde-exposed workers, but white male neoplastic,
respiratory cancer, and genitourinary cancer deaths were
analyzed by duration of employment (since being exposed
to formaldehyde). Only the PMR's for respiratory cancer
increased with increasing years of employment.
Since Marsh published this study, Infante of OSHA
has found 1 cancer of the nasal sinus and 1 naso-
pharyngeal cancer. Both men died 3 years after Marsh's
follow-up period. The worker who died of cancer of the
nasopharynx was a member of Marsh's cohort, but had been
counted as living since he had not died at that time.
16. An overlapping study was conducted by Liebling et al.
(1984). Liebling et al. identified 24 male workers who
died between January 1, 1976 and December 31, 1980
through union records, reports of former co-workers, and
a systematic review of obituaries in local newspapers.
Work histories were obtained from seniority lists.
Proportionate mortality ratios were calculated to
examine cause-specific mortality using the age, sex,
race and cause-specific mortality proportions of the
U.S. and county in which the plant is located. To
adjust for the healthy worker effect, age, sex, and
race-standardized PCMR's based on county comparisons
- 24 -
-------
were also calculated. Deaths among 18 white and 6 black
males with known HCHO exposure were identified. Race-
age-sex adjusted PMR's were significantly elevated for
cancer of the colon based on U.S., county, and county
cancer mortality proportions (PMR = 702, 424, 333,
p_<^0.05), as were PMR's for buccal and pharyngeal cancer
(PMR = 870, 952, 833, p^O.05). Liebling et al. stated
that the occurrence of a significant increase in
proportionate mortality from buccal and pharyngeal
cancer in this investigation is in accord with the type
of cancer found in HCHO-exposed rodents. Furthermore,
the authors postulated that besides nasopharyngeal
cancer, an association between HCHO exposure and cancer
of the buccal cavity and pharynx in humans is
biologically feasible since humans breathe through both
the nose and mouth, while rats and mice are obligatory
nose-breathers. Like many other studies, this study is
limited by the inability to completely separate HCHO
exposure from exposure to other chemicals.
17. Stayner et al. (1985) of NIOSH conducted a PMR study of
256 deaths among garment workers in 3 plants. Two of
these plants were included in the cohort study
identified in f!2 (Stayner, 1986). Stayner et al.
identified the 256 deaths from a death benefit fund. In
this cohort Stayner et al. observed significantly
elevated mortality from buccal cavity (PMR=750,
3 observed, p<0.001), biliary passages and liver
- 25 -
-------
(PMR=313, 4 observed, p<0.01), and other lymphatic and
hematopoietic site (PMR=400, 4 observed, p<0.05)
cancers. In analyses examining only the cancer deaths,
buccal cavity (PCMR=682) and other lymphatic and
hematopoietic site (PCMR=342) cancers remained
significantly elevated (p<0.01 and p<0.05,
respectively). Additionally, those workers with both
latency and duration of exposure of 10 years or greater
showed significantly elevated mortality from all
malignancies (PMR=137, 51 observed, p<0.05), buccal
cavity (PMR=925, 2 observed, p<0.05), biliary passages
and liver (PMR=467, 3 observed, p<0.05), and all
lymphatic/hematopoietic sites (PMR=283, 8 observed,
p<0.05), particularly other lymphatic and hematopoietic
(PMR=761, 4 observed, p<0.05) cancers.
Nonsignificant elevations in mortality were
reported for liver not specified (PMR=426, 2 observed),
skin (PMR=179, 2 observed), and all lymphatic and
hematopoietic sites (PMR=163, 10 observed), including
leukemia (PMR=400, 4 observed).
18. Delzell and Grufferman (1983) of Duke University
examined the mortality experience of 4,462 deaths
between 1976-1978 of white female textile workers.
Deaths and occupation as recorded on the death
certificates were identified from state computer
files. In this study the textile worker occupational
code included workers in industries that manufactured
- 26 -
-------
textile mill products, apparel, or other fabricated
textile products. Delzell and Grufferman observed
significant excesses in mortality from cancer of the
larynx (PMR = 280, 5 observed, p<0.05), connective
tissue (PMR = 260, 10 observed, p<0.05), cervix
(PMR = 210, 59 observed, p<0.05), other unspecified
genital organs (PMR = 270, 16 observed, p<0.05), and all
lymphopoietic sites (ICDA 200-207) (PMR = 130,
121 observed, p<0.01),• particularly non-Hodgkin's
lymphoma (PMR = 170, 51 observed, p<0.05). Decreases in
mortality that were not statistically significant were
observed for neoplasms of the lung (PMR = 90, 106 ob-
served) and of the brain (PMR = 90, 17 observed).
This study is limited by the unknown exposure
status of each death. The occupational code, textile
worker, was used as an indirect measure of formaldehyde
exposure. This study is unable to identify whether
formaldehyde had actually been an exposure, and if so,
in what concentrations. A second limitation is the
insensitivity of death certificate occupational code
analyses.
19. Fayerweather et al. (1982) of DuPont showed elevated
odds ratios, after a 15 to 24 year latency, for cancers
of the prostate (OR=4.8, 8 cases), lymphopoietic system
(OR=1.91, 6 cases), bone (OR=1.25, 3 cases), and bladder
(OR=7.0, 6 cases) among workers eligible for pension
were exposed to HCHO 5 or more years. Decreases in
- 27 -
-------
mortality were observed for those employees working 5 or
more years from colorectal (OR=0.74, 8 cases) and lung
(OR=0.79, 15 cases) neoplasms. No difference in
mortality was observed for buccal cavity neoplasms
(OR=1.0, 1 case).
Fayerweather et al. examined formaldehyde exposure
by 3 ways: by the number of years worked around
formaldehyde (less than 5 years, 5 or more years),
whether the employee had intermittent or continuous
formaldehyde exposure, and by a cumulative exposure
index. In these analyses, only the odds of mortality
from bladder and from prostatic cancer increased as the
exposure indices increased.
Fayerweather et al. did not follow those employees
ineligible for pension or those who had transferred,
potentially comprising 15 to 20% of the work group.
20. Brinton et al. (1984a) of NCI conducted a case-control
study for cancer of the nasal cavity and sinuses. They
observed nonsignificantly increased odds ratios among
males employed in the leather or shoe, chemical
manufacturing, and carpentry industries and for
exposures to chromiurn/chromates, nickel, and
insecticides/pesticides/herbicides. Among females,
increased odds ratios were observed with employment in
the textile/clothing/hosiery and paper/pulp mill
industries and for exposures to mineral oils and other
mineral/chemical gases. None of the increased odds
- 28 -
-------
ratios was significant in the presence of control for
confounding variables. Brinton et al. additionally
assessed reported occupational HCHO exposure and found
an odds ratio less than 1.0. This ratio was unstable
based on only one male and one female.
To examine the relationship between employment in
the textile and apparel industries with the risk of
nasal cancer, Brinton et al. (1984b) further analyzed
the data from their previously published case-control
study (1984a). The industries included textile and
cotton mills, apparel manufacturing, and hosiery.
Brinton et al. found an elevated risk of nasal cancers
associated with employment in the textile or apparel
industries, but the increased relative risk was found
only among female workers. When histologic types of
nasal cancer were evaluated, both males and females were
found to be at increased risk of nasal adenocarcinoma,
with further enhancement of risks with exposure to dusty
work conditions. Few individuals in this analysis had
exposure to formaldehyde (2 cases) and the ratio was
below 1.0 (OR=0.4). The authors considered this study
to provide further evidence of an association between
employment in the textile industry and risk of nasal
cancer. This study had limited ability to evaluate
nasal cavity cancer and a direct assessment of
formaldehyde exposure.
- 29 -
-------
21. Tola et al. (1980) of the Institute of Occupational
Health, Finland, conducted a case-control study for
cancer of the nose and paranasal sinsuses. Forty-five
cases were collected from the Finnish Cancer Registry
between 1970 and 1973 and were age-sex matched to
nonrespiratory cancer controls.
Analyses examining an occupational etiology showed
no single occupation being more common among the cases
than among the controls, but leisure time knitting and
sewing was significantly more common among female cases
than among female controls (OR=4.8, 19 cases). Other
factors significantly associated with the cases were
histories of serious nasal trauma, chronic rhinitis, and
sinusitis. Smoking was not significantly associated
with nasal cavity and sinus cancer.
This study is limited in its ability to evaluate
formaldehyde exposure. Occupational histories were
obtained for 69% of the cases from a next-of-kin,
potentially biasing the study towards the null
hypothesis of no-association.
22. Hernberg et al. (1983) of the Institute of Occupational
Health, Finland, conducted, with participation from
Denmark and Sweden, a collaborative case-control study
of nasal and sinonasal cancer and its possible
occupational etiology. One hundred seventy cases
diagnosed between 1977 and 1980 and reported to the
prospective cancer registries were selected. Each case
- 30 -
-------
was sex-country-age at diagnosis matched with colorectal
cancer controls.
Elevated odds ratios were observed among cabinet-
makers (OR=9.0) and mechanical engineering shop workers
(OR=2.13). Analysis for exposures showed elevated risk
with hardwood dust (OR=1.7)*, softwood dust (OR=3.4,
p<0.05)*, hardwood and softwood dust (OR=67, p<0.05)*,
welding-flame cutting-soldering (OR=2.0, 17:6,
p<0.05)**, chromium (OR=2.7, 16:6, p<0.05)**, nickel
(OR=2.4, 12:5)** electroplating (OR=1.5, 9:6)**, and
paint-lacquer (OR=3.0, 18 cases). HCHO exposures may
occur in this last category. However, wood dust
exposure is common and confounds the observed elevation.
23. Hardell et al. (1982) of Umea, Sweden conducted a case-
control study of nasal and pharyngeal cancers. Seventy-
one cases, first diagnosed between 1970-1979, and 541
referents were specifically studied for relationships
with phenoxy acid or chlordane exposure. Cases were
selected from the Swedish Cancer Register and referents
were utilized from earlier case-control studies of soft
tissue sarcoma and lymphoma and of colon neoplasms.
These referents were apparently representative of the
population between 1970 and 1978.
Hardell et al. observed increased risks with
exposure to chlorophenol (OR=6.7, 95% CI:2.8-16.2) and
*Adjusted for smoking.
**0dds ratio based on discordant pairs, concordant:
discordant pairs noted.
- 31 -
-------
phenoxy acids (OR=2.1, 95% CI:0.9-4.7). The odds ratio
for chlorophenol remained significantly elevated
(OR=6.7) when controlled for wood dust exposure. Of
interest to this review, Hardell et al. observed a
significantly increased odds ratio (OR=5.80, p<0.05)
between nasal cancer and work in particleboard
production. It is not known whether this observation is
confounded by wood dust exposure, which can occur in
particleboard production.
24. Olsen et al. (1984) of the Danish Cancer Registry
conducted a case-control study of nasal cancers. This
study examined 839 cancer registry cases (560 males,
279 females), diagnosed between the years 1970-1982, who
were matched with 2,467 controls with cancer of the
colon, rectum, prostate, and breast on age-sex-year of
diagnosis. The researchers used a nationwide data
linkage system which has linked cancer cases and
previous employment. Occupational histories came from
the National Supplementary Pension fund, established in
1964, and the Central Population Registry. Use of these
national data sets eliminated the potential for recall
bias since cases and controls were not interviewed. In
this case-control study, Olsen et al. tested for
associations between HCHO, wood dust, paint-lacquer-
glue, and metal exposure and sino-nasal cancers.
Significantly increased risks were found for nasal
cavity cancer for exposure to HCHO (OR=2.8, 95%
- 32 -
-------
CI:1.8-4.3), wood dust (OR=2.5, 95% CI:1.8-3.9), and
paint-lacquer-glue (OR=2.1, 95% CI:1.4-3.0). Exposure
to both wood dust and HCHO can occur simultaneously, and
Olsen et al. performed a stratified analysis which
controlled for wood dust exposure. In this analysis,
the elevated risk with HCHO exposure was reduced to 1.6
and became nonsignificant. In this stratified analysis,
both HCHO and wood dust exposure together resulted in an
additive risk (OR=4.1, p<0.05, 95% CI:2.3-7.3).
25. Hayes et al. (1986), formerly of the Erasmus University
of Rotterdam, presented findings of a case-control study
of nose and nasal sinus tumors at the 3rd International
Conference on Epidemiology and Occupational Health in
Dublin, Ireland. In their published study, Hayes et al.
(1986) identified factors associated with 144 cases of
nasal and nasal neoplasms diagnosed between 1978 and
1981. Living and deceased population controls were used
as the comparison group. Hayes et al. observed
significant associations between male adenocarcinoma
cases and work in furniture and cabinet making (OR=120,
90% CI:30.9-613.2) and joinery (OR=16, 90% CI:2.8-
85.3). Nonsignificantly elevated risks were reported
for all cell-types and employment in leather (OR=2.3*),
metal (OR=2.1*), and floriculture (OR=3.7, 90% CI:0.
9-18.1) industries. In addition, Hayes et al. noted
significantly increased risks between nonadenocarcinomas
*90% confidence intervals not reported.
- 33 -
-------
and paint (OR=4.1, 90% CI:1.5-11.2), benzene (OR=2.3,
90% CI:1.4-3.8), and HCHO (OR=2.2, 90% Girl.2-4.0)
exposure.
In analyses examining HCHO exposures among male
study subjects with no or low levels of wood dust
exposure, different results were observed using 2 inde-
pendent assessments, A and B, for HCHO exposure. By
classification A, a significant excess risk (OR=2.5*,
90% CI:1.2 - 5.0, 15 cases) was observed with HCHO
exposure. The authors stated there appeared to be a
dose response relationship with level of exposure. By
classification B, the odds ratio was reduced to 1.6 and
was not statistically significant (90% CI:0.9 - 2.8,
24 cases). The authors attempted analyses of this type
for males subjects with high level of wood dust exposure
using classification B data. In this analysis, the
authors observed an apparent elevation of the odds ratio
(OR=1.9, 90% CI:0.7 - 5.5, 15 cases).
EPA has analyzed the data from Classification A for
male subjects with no or low levels of wood dust
exposure and with high levels of wood dust exposure. In
this analysis, the weighted odds ratio for HCHO exposure
was significant elevated (OR=1.9, 95% CIrl.l - 3.8).
Additionally., the authors derived HCHO risks for
those classified as high wood dust exposed on both
assessments with respect to the wood dust exposure. A
statistically significant elevation (odds ratio) in
- 34 -
-------
nasal cavity cancer risk (all tumors: 3.4, 90% CI: 1.1-
10.4; squamous cell carcinoma:3.8, 90% CI: 1.1-13.0) was
observed. There was, also, a significant (p<0.05)
association for trend with time since first exposed for
all tumor types and for squamous cell carcinomas.
26. Roush et al. (1985) reported at the 1985 Meetings of the
Society for Epidemiologic Research results of a case-
control study of nasal sinus and nasopharyngeal
cancers. This study is yet to be published. At the
time of the SER presentation, 198 sinonasal and
173 nasopharyngeal cancer cases had been identified, for
the past 41 years, from the Connecticut Tumor
Registry. Controls (n=608) were sampled from death
certificates. The authors state that occupational
histories were derived from city directories and from
death certificates.
Apparent elevations in the odds ratio for combined
sinonasal and nasopharyngeal cancer (OR=2.2, 95% CI:0.7-
7.0) were reported for work in the rubber industry and
printing (OR=1.2, 95% CI:0.4-3.7), morticians (OR=2.1,
95% CI:0.7-6.5), and for physicians and dentists
(OR=1.5, 95% CI:0.5-5.1). All 4 categories were
identified as having potential formaldehyde exposure.
In analyses that examined occupational -formaldehyde
exposure, 20 or more years prior to death, no
associations were observed for those cases and controls
over 68 years of age between either sinonasal cancer or
- 35 -
-------
nasopharyngeal cancer. For those cases and controls
less than 68 years of age, apparent elevations in the
odds ratio were observed for both sinonasal cancer
(OR=1.4, 95% CI:0.6-3.0) and nasopharyngeal cancer
(OR=1.7, 95% CI:0.9-3.5).
Use of city directories and death certificates to
ascertain occupational histories lacks sensitivity and
may potentially introduce bias.
27. Partenan et al. (1985) conducted a nested case-control
study of 60 respiratory cancer cases among male
production workers in the wood industry and in
formaldehyde glue manufacturing. Respiratory cancers
were defined as mouth (other), tongue, pharynx, nasal
sinuses, larynx, and lung (trachea). Three referents
alive at the time of diagnosis of the corresponding case
were selected as controls. These referents were matched
to each case by year of birth. Smoking histories were
obtained for both cases and controls by mailed
questionnaires or by interview.
Smoking and survival status were controlled for in
the analyses since more complete work histories were
obtained for subjects who were alive at the time of data
collection. In addition, HCHO exposure was assessed
through the use of a job exposure matrix. Odds ratios
were calculated for 1) cumulative HCHO exposure >3 ppm-
months 2) cumulative HCHO exposure >_3 ppm-months, _^10
year latency, 3) peak exposure >2 ppm, 4) peak exposure.
- 36 -
-------
>2 ppm, >_10 year latency, 5) HCHO-wood dust exposure
>JL month, 6) HCHO-wood dust exposure >_l month, _>.1° years
latency, and 7) "ever exposed."
Results of the analyses showed apparent elevations
in the odds ratio with cumulative HCHO -exposure which
accounted for a 10 year latency (OR = 1.6, 8 cases) and
with "ever exposed" (OR = 1.52, 55 cases). The
exposure-response relation between HCHO exposure and
respiratory cancer was analyzed through the
classification of levels-of-exposure. In these
analyses, only the duration of exposure to HCHO-
containing wood dust suggested a positive exposure-
response relationship (1 month - 5 years; OR = 0.78,
4 cases; >5 years, OR = 1.82, 6 cases).
This study is limited by low power and a short
follow-up period; having an additional effect of
lowering the power. Thus, only very large excesses in
human risks can be ruled out.
28. Vaughan et al. (1986a,b; and as reported in SAIC, 1986)
of the Fred Hutchinson Cancer Research Center,
University of Washington, conducted a population-based
case-control study of sinonasal and pharyngeal cancers
and possible associations with HCHO. This study was
composed of 53 sinonasal cases, 27 nasopharyngeal cases,
and 205 oro-hypo-pharyngeal cases which were identified
from a tumor registry. Controls (n=557) were selected
from the general population and were matched to cases on
- 37 -
-------
sex and age. The cases were identified between 1979 and
1983, and interviews were conducted in 1983. Due to a
short survival time between diagnosis and death from
these neoplasms, next-of-kin (NOK) interviews were
obtained for 50% of the cases. Comparison of cases and
controls showed that the control population appeared to
have a higher educational level than the cases, but this
difference was not statistically significant.
Vaughan et al. examined personal, occupational, and
environmental HCHO exposures using logistic regression
analyses. Formaldehyde exposure was directly assessed
by three measures: 1) maximum exposure level, 2) number
of years in a formaldehyde job, and 3) weighted sum of
years in a formaldehyde job. Formaldehyde exposure was
indirectly assessed by identifying occupational and
domestic environments (mobile home residency,
remodeling, occupational resin-glue-adhesive exposure,
etc.) where formaldehyde had been previously reported.
Analyses showed that smoking and alcohol were
significantly associated with sinonasal cancer. When
occupational and environmental exposures were examined,
after control for smoking and alcohol consumption, the
odds ratio for exposure (>10,000 hours) to resins,
glues, and adhesives (OR = 3.8, 4 cases p_<_0.05) was
significantly elevated and the odds ratio with sawdust
exposure appeared elevated (OR =2.4, 8 cases,
p_<_0.10). The trends with increasing exposure were
- 38 -
-------
significant for both exposures. No association was
observed with the direct assessment of formaldehyde
exposure; the odds ratios were below 1.0.
In analyses of the nasopharyngeal cases, Vaughan et
al. observed significant associations with smoking and
race. After controlling for these variables,
significant elevations in the odds ratio were observed
with occupational exposure to stains, varnishes, and
solvents (OR = 4.0, p<0.05) and with ever having lived
in a mobile home (OR = 3.0, 8 cases, p<0.05), with the
highest odds ratio observed for living 10 or more years
in a mobile home (OR =5.5, 4 cases, p<0.05). Trends
for both exposures were statistically significant. An
association with formaldehyde exposure (as assessed from
the job linkage system) appeared present since the odds
ratios for formaldehyde exposure were above 1.0 and they
increased with increasing exposure, but these
conclusions are based on a total of 11 cases and were
not statistically significant.
Analyses of the oro-hypopharyngeal cases showed
significant associations with cigarette and' alcohol use,
and an alcohol-sex interaction. After considering these
variables, significant elevations of the odds ratio were
observed with exposure (>10,000 hours) with resins,
glues, and adhesives (OR = 3.9, p<0.05), stains,
varnishes, and solvents (OR = 3.9, p<0.05), and asbestos
(OR = 4.0, p<0.05). Formaldehyde exposure (as assessed
- 39 -
-------
from the job linkage system) was not significantly
associated with neoplasms of these sites. Analyses
which relied on self-reporting by eliminating next-of-
kin interviews showed an elevated odds ratio between
formaldehyde exposure, greater than 20 years, and
oropharyngeal cancer (OR = 2.0, 95% C.I.:0.9-4.6).
Mobile home residency and occupational resin, glue,
and adhesive exposure were among the exposure variables
a_ priori selected as surrogates for formaldehyde
exposure. Urea-formaldehyde resins have been used in
hardwood plywood for over 35 years (HPMA, 1984) and in
particleboard. Hardwood plywood used as prefinished
wall panels saw tremendous growth in the 1950's and
1960's, coinciding with the growth of the mobile home
industry (HPMA, 1984).
Several of the nasopharyngeal cancer cases who were
identified in this study as living in a mobile home
lived in what is generally called a recreational
vehicle. The interpretation of Vaughan's results
changes little because manufacturing practices for
mobile homes and for recreational vehicles were very
similar; formaldehyde-containing products were used in
both. By its retrospective nature, case-control study
is limited in its ability to identify if the cases had a
specific exposure and, if so, at what level of exposure.
A second limitation of this study is the accuracy
of the NOK interviews. Fifty percent of the cases were
- 40 -
-------
dead at the time of the interview and next-of-kin may
not remember all occupational histories, although one
expects that residential history may be more clearly
remembered. Inclusion of NOK interviews potentially
biases the results towards the null hypothesis of no
effect. There is support for the presence of such a
bias from analyses which eliminated NOK interviews. The
resultant odds ratios in these analyses were larger than
those odds ratios in analyses which included both live
and NOK interviews. It must be noted that the results
from NOK-eliminated analyses were not statistically
significant; a possible reflection of reduced number of
cases and, hence, reduced power. Findings which are not
statistically significant, therefore, may be due to the
third limitation of this study, reduced power to detect
very small elevations in risk.
- 41 -
-------
APPENDIX 3: ESTIMATES OF RISK USING VARIOUS
EXTRAPOLATION MODELS
-------
Appendix 3
Eleven models were used to extrapolate risks from the CUT rat
malignant tumor data. All were dichotomous ("tumor/no tumor" or
quantal) models. The formulation of each model to accommodate
quantal response data was preferred to one including time as a
variable. Simulation studies conducted for the EPA indicated that
inclusion of time as a variable would not provide much improvement
in estimation (Howe et al., 1984). Additionally, the lack of
information about causes of death of experimental animals and the
adjustments made for sacrifice data would have necessitated
assumptions that could have brought the validity of results based
on time into question.
Table 1 shows the parameter estimates (with standard errors),
log-likelihood, and X2 goodness-of-fit test statistics (with p-
values) for the eleven models: one-, three-, and five- stage models
and additive and independent background forms of the probit, logit,
Weibull, and gamma-multihit models.
Tables 2 through 6 give the maximum likelihood estimates (MLE)
and upper bound estimates for selected occupational exposures. The
shapes of most models' upper-bound estimates tend to parallel the
shapes of the models themselves, unless a procedure has been
devised to provide otherwise. This is the case for the linearized
multistage procedure, which provides a linear upper bound estimate
for a multistage model at low doses. The MLE, which is the
estimate given by a fitted model, takes only the experiment to
which the model has been fitted into account. The upper bound
estimate is intended to account for experiment to experiment
variability as well as extrapolation uncertainties.
-------
Table 1. Extrapolation Model Statistics
Model
Independent Background
Probit
Logit
Weibull
Gamma
Additive Background
Probit
Logit
Weibull
Gamma
Multistage oj qj
5-Stage 0.0 0.0
3-Stage 0.0 0.0
Parameters
(Standard Error) Loglikelihooda
a
-7.13
(0.80)
-13.61
(2.04)
-12.53
(2.00)
0.79
(0.17)
a
-14.62
(61.49)
-15.46
(25.51)
-22.67
(42.31)
0.79
(2.52)
q* i
0.0 0.
0.0 3.
P
2.85
(0.31)
5.38
(0.78)
4.75
(0.76)
10.19
(2.21)
P
4.96
(16.23)
5.95
(7.66)
7.66
(11.51)
10.19
(52.06)
» q4
7
0.00 -99.30
(0.00)
0.00 -99.31
(0.00)
0.00 -99.32
(0.00)
0.00 -99.31
(0.00)
6
6.59 -99.31
(51.33)
0.87 -99.32
(12.04)
5.31 -99.41
(20.74)
0.00 -99.30
(24.98)
qs
0 4.34E-6 1.56E-6 -99.32
46E-4 - - -104.30
1 -Stage 0.0 3.94E-2 - -
-154.17
jf* Goodness -
of-Fit
(Significance
Level)6
<0.001
(0.996)
0.008
(0.927)
0.016
(0.899)
0.001
(0.979)
0.006
(0.937)
0.019
(0.892)
0.109
(0.741)
0.001
(0.979)
0.020
(0.99)c
6.98
79.05
«.0001)c
1 The closer the loglikelihood to zero the better the model fits the
observed data.
b Significance levels less than 0.01 indicate an inadequate fit.
proximately distributed as a chi square with degrees of freedom equal
the number of doses minus the number of nonzero parameter estimates.
-------
Table 2. Risks to Mobile Home Residents (HUD Standard)
Level of Exposure: 0.15 ppm, 112 hours/week, 10 years
Point Estimate 95% Upper Confidence
Model of Added Risk Limit on Added Risk
Independent Background
Probit 0.0 0.0
Logit 2 x 10"n 2 x 10~1§
Weibull 2 x 10~w 2 x 10~9
Gamma 3 x 10~17 2 x 10~16
Additive Background
Probit 3 x 10~s 2 x 10~e
Logit 7 x 10~8 4 x 10~a
Weibull 6 x 10"6 5 x 10~8
Gamma 3 x 10~17 8 x 10~14
5-Stage Multistage 1 x 10~9 2 x 10~4
3-Stage Multistage 6 x 10~4 1 x 10~4
1-Stage Multistage 3 x 10~8 4 x 10~3
-------
Table 3. Risks to Manufacturers of Apparel (OSHA Standard)
Level of Exposure: 3.0 ppm, 36 hours/week, 40 years
Point Estimate 95% Upper Confidence
Model of Added Risk Limit on Added Risk
Independent Background
Probit 2 x 10~» 9 x 10'5
Logit 3 x 10~4 9 x 10~4
Weibull 5 x 10~4 1 x 10~8
Gamma 1 x 10~4 4 x 10~4
Additive Background
Probit 2 x 10~4 3 x 10~8
Logit 4 x 10~4 3 x 10~8
Weibull 1 x 10~3 3 x 10~3
Gamma 1 x 10~4 2 x 10~3
5-Stage Multistage 5 x 10"4 6 x 10~8
3-Stage Multistage 6 x 10~8 9 x 10~8
1-Stage Multistage 8 x 10~2 9 x 10~2
-------
Table 4. Risks to Manufacturers of Apparel
(Personal Sample)
Level of Exposure: 0.64 ppm, 36 hours/week, 40 years
Point Estimate 95% Upper Confidence
Model of Added Risk Limit on Added Risk
Independent Background
Probit 8 x 10~17 3 x 10~"
Logit 8 x 10"1 4 x 10"7
Weibull 3 x 10~7 1 x 10~6
Gamma 7 x 10~" 6 x 10~lf
Additive Background
Probit 5 x 10~7 2 x 10~8
Logit 1 x 10~fl 6 x 10~5
Weibull 5 x 10~5 3 x 10~4
Gamma 7 x 10~n 3 x 10"8
5-Stage Multistage 6 x 10~7 l x 10~s
3-Stage Multistage 6 x 10~s 7 x 10~4
1-Stage Multistage 2 x 10~2 2 x 10~2
-------
Table 5. Risks to Manufacturers of Apparel
(Area Sample)
Level of Exposure: 0.23 ppro, 36 hours/week, 40 years
Point Estimate 95% Upper Confidence
Model of Added Risk Limit on Added Risk
Independent Background
Probit 0.0 1 x 10~2B
Logit 3 x. 10~1§ 2 x 10~9
Weibull 2 x 10~8 1 x 10~8
Gamma 3 x 10~18 4 x 10~14
Additive Background
Probit 7 x 10~fl 4 x 10~6
Logit 2 x 10~7 1 x 10"5
Weibull 1 x 10~6 1 x 10~4
Gamma 3 x 10~15 4 x 10~12
5-Stage Multistage 9 x 10~9 4 x 10~4
3-Stage Multistage 3 x 10~6 2 x 10~4
1-Stage Multistage 6 x 10~3 7 x 10~s
-------
Table 6. Risks to Manufacturers of Apparel
(NIOSH Data)
Level of Exposure: 0.17 ppm/ 36 hours/week, 40 years
Point Estimate 95% Upper Confidence
Model of Added Risk Limit on Added Risk
Independent Background
Probit 0.0 0.0
Logit 6 X 10"" 4 x 10~11
Weibull 5 x 10"11 4 x 10"'
Gamma 1 x 10~16 2 x 10~«
Additive Background
Probit 4 x 10~8 3 x 10"6
Logit 1 x 10~7 7 x 10"fl
Weibull 9 x 10~6 8 x 10~s
Gamma 1 x 10~16 3 x 10"13
5-Stage Multistage 3 x 10~' 3 x 10~4
3-Stage Multistage 1 x 10~6 2 x 10~4
1-Stage Multistage 5 x 10~* 5 x 10"8
-------
APPENDIX 4: DOCUMENTATION OF HIGH TO LOW DOSE
EXTRAPOLATION MODELS USED IN
QUANTITATIVE RISK ASSESSMENT-CONCISE
DESCRIPTION
-------
TABLE OP COHTKHTS
Page
1.0 INTRODUCTION 1
2.0 GENERAL INFORMATION 2
2.1 THE NEED FOR HIGH- TO LOW-DOSE EXTRAPOLATION 2
2.2 TYPE OF DATA 2
i
2.3 TYPES OF MODELS FOR QUANTAL RESPONSE DATA 3
2.4 SPONTANEOUS BACKGROUND RESPONSE 4
3.0 PROBIT MODEL (LOGNORMAL) 8
4.0 LOGIT MODEL (LOG LOGISTIC MODEL) 10
5.0 WEIBULL MODEL (EXTREME VALUE) 11
6.0 ONE-HIT MODEL (LINEAR MODEL) 12
7.0 GAMMA-MULTIHIT MODEL 14
8.0 MULTISTAGE MODEL 17
REFERENCES 18
-------
1.0 INTRODUCTION
The Design and Development Branch (DDE) of the Exposure
Evaluation Division (BED) of the U.S. Environmental Protection
Agency currently uses six extrapolation models to estimate car-
cinogenic risk in humans from animal test data. This report pro-
vides general introductory material and a concise description of
each model suitable for insertion into the background and methods
section of a quantitative risk assessment. No mathematical equa-
tions are included and technical terms are avoided as much as
possible. The introductory material covers the reason for high-
to low-dose extrapolation, the type of data used, classes of
models and incorporation of spontaneous background response. The
six models that are described are the probit, logit, Weibull,-
one-hit, gamma-multihit and multistage models. Technical details
and a more comprehensive bibliography appear in a companion re-
port (Chesson and Zanetos, 1983).
-------
2.0 GENERAL INFORMATION
2.1 THE NEED FOR HIGH- TO LOW-DOSE EXTRAPOLATION
The most readily available source of information for
determining the health effects of toxic agents comes from experi-
mental tests on laboratory animals. In order to obtain observ-
able effects within a reasonable time the laboratory animals must
be exposed to concentrations, or doses, of the toxic substance
that are higher than those expected to be experienced by humans.
Therefore it is necessary to predict the effects at low doses
from the effects observed at higher doses. This process is
called "high- to low-dose extrapolation" and is carried out by
fitting a mathematical model to the observed data and using the
model to estimate the effect of the substance at low doses.
2.2 TYPE OF DATA
A typical experiment to collect data for high- to low-
dose extrapolation consists of several groups of animals. Each
group is exposed to a different dose, or concentration d, of the
agent under test and the numbers of animals in each group that
show a particular response within a fixed time period are record-
ed. This type of data is often called "quantal" response or "di-
chotomous" data. The data provide the basis for fitting dose
response models and extrapolating to low doses. As used here, a
dose response model is a mathematical relationship between the
applied dose d, and the proportion of animals in the group, P(d),
that show the response. When additional information, such as the
time between the start of exposure to the agent and appearance of
-------
a tumour, is available then other types of dose response models
may also be used.
2.3 TYPES OF MODELS FOR QUARTAL RBSPOSSB DATA
The mathematical models used for describing guantal
responses fall into two broad types:
• Tolerance distribution models
• stochastic or mechanistic models.
(Krewski and Van Ryzin, 1981; Brown, 1983).
Tolerance distribution models are based on the idea
that each individual in a population has its own tolerance to the
test agent. If a dose does not exceed the tolerance of an
individual then there will be no response by that individual. If
the dose exceeds the tolerance then a response will be observed.
Tolerance models differ from each other in the particular
mathematical expression used to describe the distribution of
tolerances in the population (Chand and Hoel, 1974). The dis-
tributions are generally chosen because of their descriptive
power rather than on the basis of biological processes.
Stochastic, or mechanistic, models are derived from
plausible biological arguments which lead to an expression for
P(d), the expected proportion of animals that will show a
response at dose d. Sometimes the mechanistic argument leads to
a tolerance distribution. Thus the distinction between the two
types of model is not always very obvious.
-------
The behavior of a model at low doses is of particular
importance since it is the low dose region about which predic-
tions are to be made. In this region the shape of the dose
response curve predicted by a model may be convex, linear or con-
cave (Figure 1). The usual procedure is to estimate the dose d*f
corresponding to a particular low level of risk having taken into
account spontaneous background response (see Section 2.4 below).
The excess risk is commonly set at 10-6, i.e. the level at which
the expected proportion of individuals that will show the re-
sponse as a result of exposure to dose d* is one in 1,000,000.
The dose d* is called the "virtually safe dose" (VSD). Figure 1
shows how the estimate of the VSD depends on the shape of the
dose response curve. A linear curve will give a lower VSD than a
comparable convex curve and a higher VSD than a concave curve.
These relationships are used in some extrapolation procedures
that attempt to place a lower bound on the VSD rather than obtain
an actual estimate of it. The estimates obtained in this way are
considered 'conservative1 in that they are expected to
overestimate risk and underestimate the VSD.
2.4 SPONTANEOUS BACKGROUND RESPONSE
In an experiment to determine a dose response curve it
is possible that some animals might show a response even though
they receive zero dose of the chemical. This spontaneous back-
ground response may be a result of many factors including the
-------
FIGURE 1. THE RELATIONSHIP BETWEEN THE SHAPE OF THE DOSE RESPONSE
CURVE AT LOW DOSES AND THE VSD (d*) (ASSUMING NO
SPONTANEOUS BACKGROUND RESPONSE)
(a) Concave
(b) Linear
(c) Convex
-------
presence of another response causing chemical, the genetic make-
up of the strain of animal, or a background level of the toxic
substance which is present in the environment. The method of
incorporating the spontaneous background response into the model
affects the shape of the dose response curve at low doses and
hence estimates of the VSD.
If the background response is assumed to be totally
independent of the response to the experimental dose then the
shape of the dose response curve remains qualitatively the same
as when no background response is included. However, if the
background response is assumed to be additive in the sense that
the effective dose is taken to be the background dose plus the
experimental dose then in many cases the dose response curve be-
comes linear at low doses (Crump e_t aJL, 1976). This is true for
most of the commonly used models, including the probit, logit,
Weibull, one-hit, gamma-multihit and multistage models. A dose
response curve that is linear at low doses will result in a
smaller VSD than a comparable dose response curve that is convex
at low doses. Thus, a model that includes additive background
effects is likely to give estimates of the VSD which are smaller
than those estimated by a model without additive background.
A model may include both independent and additive back-
ground but as long as the additive component is non-zero the dose
response curve at low doses will be linear (Crump et al, 1976;
Peto, 1978).
It is often difficult to decide from experimental data
whether background responses are independent, additive, or both
-------
(Brown, 1993). Therefore, it has been suggested that models in-
corporating additive background be used unless there is good
evidence to assume otherwise (State of California, Health and
Welfare Agency, 1982).
-------
3.0 PROBIT MODEL (LOGNORNAL)
The probit model is a tolerance distribution model in
which the distribution of individual tolerances is assumed to
follow a lognormal distribution. The dose response curve is s-
shaped and symmetric about the 50 percent level. Since experi-
mental data often show this sort of pattern, the probit
model has been used extensively in toxicological studies and a
t
standard statistical theory has been developed around it (Finney,
1971). The use of the probit model to extrapolate to low doses
is more recent.
When the background response is independent of the in-
duced response the probit model approaches zero very rapidly as
dose decreases, more rapidly than other common models used in
low-dose extrapolation (Krewski and Van Ryzin, 1981). This means
that the probit model with independent background cannot be
linear at low doses and tends to give lower estimates of risk and
higher VSD's than those obtained from other models, or from
linear extrapolation.
The Mantel-Bryan procedure (Mantel and Bryan, 1961;
Mantel e_t al, 1975) uses the probit model to obtain a "conserva-
tive" estimate of the VSD. By taking an arbitrary value of 1 for
the slope parameter of the model and extrapolating from the range
of observable responses it is assumed that the extrapolated curve
will lie above the true curve and therefore provide a conserva-
tive estimate of a VSD. However the validity of this procedure
has been questioned (Crump, 1977). In particular, it is not
-------
clear that taking the value of the slope parameter as 1 is neces-
sarily conservative (Connfield et al, 1978).
-------
10
4.0 LOGIT MODEL (LOG LOGISTIC MODEL)
The logit model is a tolerance distribution model which
has a very similar appearance to the probit model within the
range of observable responses. It is S-shaped and symmetric
about the 50 percent response level. However it differs from the
probit model in that at low doses it approaches zero much more
slowly. The model can be derived from chemical kinetic theory
and wa's proposed as a dose response model by Worcester and Wilson
(1943) and Berkson (1944).
With independent background the excess risk may be
linear, convex or concave at low doses. Low-dose linearity
implies a concave dose response curve at higher doses. With
additive background the excess risk will always be linear at low
doses (Peto, 1978). Therefore, linear extrapolation procedures
will tend to give estimates of VSD that are close to or smaller
than those based on the model itself unless the dose response
curve is concave at low doses. In fitting a variety of models to
20 data sets Krewski and Van Ryzin (1981) found that estimates of
VSD based on the logit model were smaller than those based on the
probit model and similar to those based on the gamma.-multihit
model.
-------
11
5.0 WEIBDLL MODEL (EXTREME VALUE)
The Weibull model is a tolerance distribution model
which is suggested by human cancer incidence patterns (Cooke e_t
al, 1979). Pike (1966) showed that two quite general assump-
tions, (1) cancer begins in a single cell, and (2) individual
cells behave independently, can lead to a Weibull distribution
for tolerances. The model can also be derived from a •time-to-
tumour* argument (Peto et al, 1972) or from a model based on
critical cell clusters (Scott and Hahn, 1980). Generalized forms
of the Weibull model are discussed by Carlborg (1981a).
With independent background the excess risk at low
doses predicted by the Weibull model behaves in the same way as
the logit model. The excess risk may be linear, convex or con-
cave depending on the value of the shape parameter. Low dose
linearity implies a concave dose response curve at moderate and
high doses, with additive background, the excess risk will al-
ways be linear at low doses (Peto, 1978).
Estimates of the VSD based on the Weibull model tend to
be less conservative than the multistage model but more conserva-
tive than the probit, logit and gamma-multihit models (Krewski
and Van Ryzin, 1981).
-------
12
6.0 ONE-BIT MODEL (LINEAR MODEL)
The one-hit model is derived from a mechanistic de-
scription of the carcinogenic process. Suppose that there is a
response after a susceptible site has been •hit" by a single bio-
logically effective unit of dose within a fixed period of time.
By assuming that the number of hits over that time period follows
a Poisson distribution and the average number of hits is propor-
tional to the dose, a formula is easily obtained for the proba-
bility (P(d)) of obtaining a response at a given dose (d). The
Poisson distribution assumption is appropriate when hits occur
randomly through time and the occurrence of a hit has no effect
on the occurrence of other hits.
In the absence of spontaneous background responses the
one-hit model has only one unknown parameter and is always linear
at low doses and concave at moderate and high doses. Because of
its low dose linearity, it is often referred to as the linear
model. It is a special case of the gamma-multihit, multistage
and Weibull models.
The one-hit model does not provide a good fit to many
sets of empirical data because the model is concave at higher
dose levels whereas many data sets are convex. It appears to be
appropriate for only one (hexachlorobenzene) of the 20 data sets
considered by Krewski and Van Ryzin (1981). However, it has been
used extensively in low dose extrapolation as a conservative
estimate of risk, assuming that the true dose response curve is
likely to be convex at low doses and hence lie below that of the
one-hit model (Hoel e_t al, 1975; BEIR Report, 1972). In some
-------
13
situations the one-hit model is only fitted to the lowest dose
groups where the experimental data are consistent with the model
(Altshuler, 1976). The model has been criticized as being unduly
conservative in some circumstances. For example, Van Ryzin and
Rai (1980) found that for ethylene thiourea the VSD (for risk
level 10~6) estimated by the one-hit model was approximately
1/60,600th that of the multihit model and l/8050th that of the
multistage model.
-------
14
7.0 GAMMA-HOLTIHIT MODEL
The gamma-milltihit model can be derived from a mechan-
istic description of the carcinogenic process. Suppose the
response to a particular chemical is the result of k biological
precursor events or "hits11 at a susceptible site within a fixed
period of time. By assuming that the number of hits over that
time period follows a Poisson distribution and the average number
of hits is proprotional to the dose, a formula can be obtained
for the probability (P(d)) of obtaining a response at a given
dose (d). The Poisson distribution assumption is appropriate
when hits occur randomly through time and the occurrence of a hit
has no effect on the occurrence of other hits. When only one hit
(k«l) is required to cause a response, the model is referred to
as the one-hit model.
The gamma-multihit model can also be regarded as a
tolerance distribution model since the formula for P(d) is
identical to the dose response curve generated by assuming that
each individual in the population has a particular tolerance
level to the chemical and that the distribution of tolerance
levels follows a gamma distribution. In this situation the gamma
distribution is merely used to describe the shape of the dose
response curve and has no mechanistic implications.
For small doses the dose response curve is concave when
k < 1, linear if k-1 and convex if k > 1 (Figure 1). Thus the
gamma-multihit model provides a greater variety of behavior at
low doses than models which can have only linear or convex dose
response curves at low doses. However, k < 1 is not easily
-------
15
interpretable in terms of the mechanistic "hit" model, and one
has to resort to the descriptive tolerance distribution
interpretation in this case.
Although the gamma-multihit model has been recommended
for use in risk assessment calculations (Food Safety Council,
1978), this recommendation has been criticized by Baseman et aJL
(1981) because the model can produce estimates of the VSD which
are unrealistically high or unrealistically low in certain
situations. These problems appear to be less likely to occur
when an additive background effect is included in the model
thereby causing the dose response curve to be linear at low
doses.
The gamma-multihit model tends to produce estimates of
the VSD which are less conservative than the multistage and
Heibull models and more conservative than the logit and probit
models (Krewski and Van Ryzin, 1981).
-------
16
8.0 MULTISTAGE MODEL
The multistage model was derived to account for the
observation that for many types of cancer, death rate is propor-
tional to some power of age (e.g. Nordling, 1953). The model was
developed by Armitage and Doll (1961) and extended by Crump et al
(1976). The Armitage and Doll model assumes that a cell line -
goes through k distinct stages in a specific order before becom-
ing cancerous and that the rate at which it progresses through
the ith stage is a constant \±. Different cell lines develop
independently and the time to cancer is determined by the most
rapidly developing line. This model predicts that cancer inci-
dence will increase as (age)*'1.
Crump et al (1976) extended the Armitage and Doll model
by assuming that Xi, the rate at which a cell goes through the
ith stage, is linearly related to dose. From these assumptions,
the probability of a response, P(d), at dose d can be expressed
in terms of a polynomial in d. The model is essentially
unchanged irrespective of whether indepenedent or addititve
background responses are assumed. It is linear at low doses if
the polynomial in d has a linear term and convex otherwise. It
cannot be concave at low doses, but it can be simultaneously
linear at low doses and convex at moderate doses. When k«l,
i.e., when there is only one stage, the model reduces to the one-
hit model.
The multistage model cannot describe concave dose
response curves such as those observed with DDT, vinyl chloride,
diethylstilbestrol and ethylene dibromide (Carlborg, 1981b).
-------
17
Also, the shape of the curve in the extrapolated low dose region
is relatively insensitive to the shape of the dose response func-
tion in the observable range (Carlborg, 1981a).
The parameters of the multistage model are more compli-
cated to estimate than for other models. However, computer pro-
grams are available to carry out the calculations (Crump and
Watson, 1979; Howe and Crump, 1982).
»
Crump (1982) proposed a method of linear extrapolation
based on the multistage model, which has been adopted by the EPA
in setting water quality criteria (USEPA, 1980) and in other
areas of risk assessment. The method is an improvement of the
extrapolation procedure proposed by Crump e_t al (1977). It
involves approximating the dose response curve by a straight line
with slope given by the estimated linear term of the multistage
model. This modification has been referred to as the "lineariz-
ed* multistage model and has an advantage over some other linear
extrapolation procedures in that information from all the experi-
mental dose groups is used, not just the lower groups. The
6LOBAL79 computer program (Crump and Watson, 1979) calculates
confidence limits for the linearized multistage model. A newer
program (GLOBAL82, Howe and Crump, 1982) uses a different method
to allow valid confidence limits for estimates at higher doses as
well as at low doses. At low doses the results obtained from
GLOBAL?9 and GLOBAL82 should be very similar.
-------
18
REFERENCES
Altshuler B. 1976. A Bayesian approach to assessing population
risks from environmental carcinogens. In: Environmental Health.
A. Whittemore, ed. SIAM: Philadelphia
Armitage P, Doll R. 1961. Stochastic models for carcinogenesis.
Proceedings 4th Berkeley Symposium IV, pp. 19-38.
Berkson J. 1944. Application of the logistic function to
bioassay. J. American statistical Association 39:134-167.
BEIR Report. 1972. The effects on populations of exposures to
low levels of ionizing radiation. Report of the advisory
committee of the biological effects of ionizing radiations.
National Academy of Sciences. National Research Council,
Washington, DC. Government Printing Publication No. 0-489-797.
Brown CC. 1983. Learning about toxicity in humans from studies
on animals. Cheratech, pp. 350-358.
Carlborg FW. 1981a. 2-Acetylaminofluorene and the Weibull
model. Fd Co sine t Toxicol 19:367-371.
Carlborg FW. 1981b. Multistage dose-response models in carcino-
genesis. Fd Cosmet Toxicol 19:361-365.
Chand N, Hoel DG. 1974. A comparison of models for determining
safe levels of environmental agents. In: Reliability and
Biometry; Statistical Analysis of Lifelength. SIAM:
Philadelphia, pp. 681-700.
Chesson J, Zanetos MA. 1983. Documentation of high- to low-dose
extrapolation models used in quantitative risk assessment. Draft
Report. Washington, DC: Office of pesticides and Toxic
Substances. U.S. Environmental Protection Agency. Contract 68-
01-6721.
Cook PJ, Doll R, Fellingham SA. 1969. A mathematical model for
the age distribution of cancer in man. Int J Cancer 4:93-112.
-------
19
BS
(Continued)
Cornfield J, Carlborg FW, Van Ryzin J. 1978. Setting tolerances
on the basis of mathematical treatment of dose-response data ex-
trapolated to low doses. In: proceedings of the First Interna-
tional Congress on Toxicology. Plaa GL, Duncan WAM, eds. New
York: Academic, pp. 143-164.
Crump KS. 1977. Theoretical problems in the modified Mantel-
Bryan procedure. Biometrics 33x752-755.
i
Crump RS. 1982. An improved procedure for low-dose carcinogenic
risk assessment from animal data. J. Environmental Path
Toxicology 5(2):675-684.
Crump KS, Guess HA, Deal KL. 1977. Confidence intervals and
tests of hypothesis concerning dose-response relations inferred
from animal carcinogenicity data. Biometrics 33:437-451.
Crump RSr Hoel DG, Langley CH, Peto R. 1976. Fundamental car-
cinogenic processes and their implications for low dose risk
assessment. Cancer Research 36:2973-2979.
Crump RS, Watson WW. 1979. GLOBAL79: A FORTRAN program to
extrapolate dichotomous animal carcinogenicity to low dose.
Finney DJ. 1971. Probit analysis (3rd edition). London:
Cambridge University Press.
Food Safety Council. 1978. Proposed system _for food safety
assessment. Fd Cosraet Toxicol 16 Supp 2:1-136.(Revised report
published June 1980, by the Food Safety Council, Washington, DC.)
Baseman JR, Hoel DG, Jennrich RI. 1981. Some practical problems
arising from use of the gamma multi-hit model for risk estima-
tion. J. Toxicol Environmental Health 8:379-386.
Hoel DG, Gaylor DW, Rirschstein RL, Saffiotti U, Schneiderman MA.
'1975. Estimation of risks of irreversible, delayed toxicity. J.
Toxicol and Environmental Health 1:133-151.
-------
20
REFERENCES
(Continued)
Howe RB, Crump KS. 1982. GLOBAL 82. A computer program to ex-
trapolate guantal animal toxicity data to low doses. Report to
Office of Carcinogen Standards Occupational Safety and Health
Administration, US Dept. of Labor, Contract 41 DSC 252C3.
Krewski D, Van Ryzin J. 1981. Dose response models for quantal
response toxicity data. In: Current Topics in Probability and
Statistics. Csargo M, Davson D, Rao JNK, Saleh E, eds. North-
Holland: New York, pp. 201-281.
Mantel N, Bryan WR. 1961. "Safety" testing of carcinogenic
agents. J National Cancer Institute 27(2):455-470.
Mantel N, Bohidar NR, Brown CC, Ciminera JL, Tukey JW. 1975. An
improved Mantel-Bryan procedure for "safety" testing of carcino-
gen's. Cancer Research 34:865-872.
Peto R. 1978. Carcinogenic effects of chronic exposure to very
low levels of toxic substances. Environmental Health Perspec-
tives 22:155-159.
Peto R, Lee PN, Paige WS. 1972. Statistical analysis of the
bioassay of continuous carcinogens. Br. J. Cancer 26:258-261.
Pike MC. 1966. A method of analysis of a certain class of ex-
periments in carcinogenesis. Biometrics 22:142-161.
Scott BR, Hahn pp. 1980. A model that leads to the Weibull
distribution function to characterize early radiation
probabilities. Health Physics 39:521-530.
State of California Health and Welfare Agency. 1982. Carcinogen
identification policy: a statement of science as a basis of
policy. Section 2: Methods for estimating cancer risks from
exposures to carcinogens.
US EPA. 1980. (U.S. Environmental Protection Agency). Hater
quality criteria documents: availability. Federal Register
45(231):79347-79357.
-------
21
REFERENCES
(continued)
VanvRyzin J, Rai K. 1980. The use of guantal response data to
make predictions. In: The Scientific Basis of Toxicity
Assessment. Witschi H, ed. Elsevier/North Holland: Biomedical
Press.
Worcester J, Wilson EB. 1943. The determination of LD50 and its
sampling error in biossay. Proceedings of National Academy of
Sciences 29:79-85.
-------
APPENDIX 5: SENSITIVITY ANALYSIS OF CUT RAT DATA
USING THE LINEARIZED MULTISTAGE MODEL
-------
Appendix 5
Ten perturbations of the final CUT formaldehyde combined male
and female Fischer 344 rat data were constructed in 3 ways: one, as
if a dose had not been run; two, as if the response had been
different at an intermediate dose; or, three, the response was
different at the highest dose or control (see Table 1).
Perturbations 1 and 2 removed one intermediate dose and also
increased the number of animals at risk by 1, thereby lowering the
response rate at the highest dose. Perturbation 3 eliminated the
response and number of animals at risk at the high dose entirely.
Perturbation 4 gave the CUT data set a positive response at
control. Pertubations 5 and 7 use the correct final denominator or
number of animals at risk at the highest dose, but vary the numbers
responding. Perturbations 6 and 8 lower or raise the response at
the next to the highest dose. Perturbation 9 eliminates the control
data entirely. Perturbation 10 gives a positive response at the
lowest positive dose where one did not appear in the CUT data set.
For all perturbations, with a five-stage model the fit was
adequate (see Table 2, first column). While the individual maximum
likelihood coefficients varied from perturbation to perturbation,
the predicted upper bound risks varied by less than a.factor of 2
(see Table 3).
A graphical representation of the 10 perturbations is given in
Figures 1 an 2 which, upon inspection,.reveal just how similar the
"curves" are. Note in Figure 2 that the curve for perturbation 2
has a considerably steeper slope and is rather an outline. This is
because the lowest positive dose point was dropped, a section of the
curve where the most information is needed.
-------
Table 1. Sensitivity of the 5-Stage Model
(Model Statistics)
Dose (ppm)
Responding/Tested
Parameter Estimates
Perturbation
0.0
2.0
5.6
14.3
q»
qs
1
2
3
4
5
6
7
8
9
10
0/156
0/156
0/156
1/156
0/156
0/156
0/156
0/156
0/156
0/159 - 94/141
2/150 94/141
0/159 2/150
0.0 0.0
0.0 1. 5E-S
0.0 0.0
0/159 2/153 94/140 3.16E-1 0.0
0/159 2/153 95/140 0.0 0.0
0/159 1/153 94/140
0/159 2/153 93/140
i
0/159 3/153 94/140
0/159 2/153 94/140
1/159 2/153 94/140
0.0 0.0
0.0 0.0
0.0 0.0
0.0 0.0
0.0 1.68E-3
0.0 0.0
5.99E-7 8.00E-7
0.0 0.0
0.0 0.0
0.0 0.0
0.0 0.0
0.0 0.0
4.06E-1I 0.0
0.0 0.0
0.0 0.0
0.0 1.84E-0
5.16E-* 1.47E-6
0.0 2.42E-8
0.0 1.86E-8
4.00E-6 1.62E-0
0.0 1.85E-6
4.67E-I 1.50E-6
1.52E-S 8.0 IE-7
4.34E-I 1.56E-6
,0.0 1.81E-6
-------
Table 2. Sensitivity of the 5-Stage Model
(Model Goodness-of-Fit)
Perturbation
1
2
3
4
5
6
7
8
9
10
X1 Goodness-of-Fit
(Significance Level)
9 . 35E-3
(7.995)«
1.69E-26
(1.000)
1.24E-2
(7.995)
1.04
(7.75)
1.96E-2
«.50)
0.21
(0.98)
2.07E-2
(0.99)
4 . 44E-2
(7.90)
2.0 IE- 2
(7.90)
0.69
(7.98)
Loglikelihood
-89.758
-100.37
-10.634
-106.065
-98.592
-94.794
-100.023
-103.452
-99.34
-105.697
Approximately distributed as a chi square with degrees of
freedom equal to the number of doses minus the number of
nonzero parameter estimates.
-------
Table 3. Coefficients (q*'s) corresponding to the upper bound on risk for the 5-Stage
Model for ten perturbations of the model.
Perturbation
Number q*c
1 0.0000
2 0.0000
3 0.0000
4 2.5652 X ID'3
5 0.0000
6 0.0000
7 0.0000
8 0.0000
9 0.0000
10 0.0000
q
4.2556
4.6685
3.6499
2.3843
2.6652
1.7512
2.7074
3.5135
2.6866
4.9145
*l
X 10'3
X lO-3
X lO'3
x 10'3
X lO'3
X lO'3
X lO'3
x lO"3
X lO'3
X 10'3
0
0
0
0
0
0
0
0
0
0
q*2
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
0
0
0
0
0
0
0
0
0
0
q*3
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
0
0
0
0
0
0
0
0
0
0
q*4
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
1
1
1
1
1
1
1
1
\
q*5
.7-353 X
.7056 X
0.0000
.7832 x
.8176 X
.7966 x
.7450 X
.7660 X
.7809 X
.7220 X
10'6
10'a
lO'6
io-e
10-'
io-«
io-«
io-°
10-"
-------
Figure 1. Sensitivity of the 5-Stage Model for all ten perturbations of the dose-response
data for Fischer 344 rats.
B.<1F-04
7.6E-04
6.7FI-04-
A
O 5.9F-04
0
I
T 5.0E-04
I
0
N 4.2E-04-
A
L
3.4E-04-
R
I
S 2.5E-04-
K
1.7E-04
8.4E-05
O.OE-»00
0
0
0
0
0
0
0
0
o
0
0
0
0
o
0
0
0
0
o
0
0
0 l
0
0
0°
T 1 1 1 1 T 1 1 i 1 r
.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.6 O.9 1.
Legend: 111 Perturbation 1
222 Perturbation 2
333 Perturbation 3
F/POSUHF I. rvFL (PPMI
444 Perturbation 4 777 Perturbation 7
555 Perturbation 5 888 Perturbation 8
666 Perturbation 6 999 Perturbation 9
000 Perturbation 10
-------
Figure 2. Sensitivity of the 5-Stage Model rescaled to display perturbations 1-9 of the
dose-response data for Fischer 344 rats.
A
0
0
I
T
O
N
A
L
R
I
S
K
.
1
9
8.
7.
B.
5.
4.
3.
2.
1 .
0.
OF-05
OE-06
OF-OB
OE-06
OE-06
OE-OB
OF-06
OE-06
OE-06-
OE-06
OF+OO
1
0
1
2' •
2
2 8
*" Q
2 8
2
2 8
2 a
2 B
9 7
« 2 8 -,9
2 Z5
2 8 Z5
22 8 7§
2 8 25
2 7 8
2 8 §5
2 8 2 4 D
22 8 §*
S— 9 -t
d9 33I
2 8 al 3.1
2 8 a!' ^3t*
2 B flB •« 3 • ^
2 88a a|JI^ 3ii**$
i 1 1 1 1 1 1 1 1 1 r
0 O.t 0.2 0.3 0./1 0.5 0.6 O.7 0.8 O.9 l.{
Fxposunr ir-vFL (PPM)
Legend: 111 Perturbation 1 444 Perturbation 4 777 Perturbation 7
222 Perturbation 2 555 Perturbation 5 888 Perturbation 8
333 Perturbation 3 666 Perturbation 6 999 Perturbation 9
------- |