QUANTITATIVE RISK ASSESSMENT
FOR COMMUNITY EXPOSURE
TO VINYL CHLORIDE
Arnold M. Kuzmack
Office of Planning and Evaluation
U. S. Environmental Protection Agency
Washington, D. C. 20460
and
Robert E. McGaughy
Office of Health and Ecological Effects
U. S. Environmental Protection Agency
Washington, 0. C. 20460
December 5, 1975
-------
EXECUTIVE SUMMARY
Vinyl chloride has produced liver angiosarcoma, a very rare form
of cancer, as well as other cancers and serious liver damage in
occupationally exposed populations. Since it is generally considered
prudent to assume that there is no threshold for chemical carcinogens,
the small concentrations of vinyl chloride observed in the ambient air
in the vicinity of plants producing vinyl chloride monomer (VCM) or poly-
merizing it to polyvinyl chloride (PVC) involve some public health risk.
This paper reports the results of an analysis to estimate quantitatively
the risk resulting from vinyl chloride emissions and assess the re-
liability of the estimates. In addition, a search was initiated of the
residence of people who have died of liver angiosarcoma in order to
detect evidence of adverse effects among people living near plants.
The analysis involves three steps: estimation of the number of
people exposed, the concentrations of vinyl chloride to which they are
exposed, and the number of cancers and other health effects resulting
from this exposure. Of these, the last is by far the most difficult
to make and involves the most conceptual problems.
The number of people living at various distances from each VCM
and PVC plant was determined by an analysis of census data. All told,
some 4.6 million people live within 5 miles of these plants.
Ambient concentrations of vinyl chloride in the vicinity of plants
were determined using standard methods of diffusion modeling. Exposures
at each location were adjusted to allow for the types, numbers and sizes
of plants present and the meteorological conditions which affect the
dispersion of pollutants. It was found that the 4.6 million people
living within 5 miles of uncontrolled VCM or PVC plants are exposed to
an average concentration of about 17 parts per billion.
Quantitative estimates of health effects likely to result from this
exposure were made by predicting from animal data the probability that
highly-exposed workers would get angiosarcoma, checking that prediction
by calculating the same probability using epidemiological studies of
workers, and then projecting the results to ambient concentrations around
plants. Cancers other than liver angiosarcoma were assumed to occur at
the same relative rate as at the higher doses. Birth defects have been
reported in communities with VC and VCM plants. These effects are not
considered in detail here because quantitative estimates of these
effects can not be made at this time.
Projections from incidence rates in rats to incidence rates in
highly-exposed workers were made by assuming that a lifetime exposure
to rats would produce the same number of effects as a lifetime of
exposure to humans, which means roughly that one year of exposure to
rats yields the same incidence rate as 30 years of exposure to people.
-------
ii
It was found, using a linear dose-response model extrapolating to
zero, that the probability of getting angiosarcoma due to a year of
continuous exposure is 71 cases per year of continyous exposure per
ppm of vinyl chloride per million people exposed. The corresponding
probability of getting cancer of all sites is 150 cases per year of
continuous exposure per per ppm of vinyl chloride per million people. On
this basis the probability of getting liver angiosarcoma in workers
exposed to 350 ppm for 7 hours per day, 5 days per week would be 0.0052
per person per year of exposure.
From epidemiological studies of exposed workers, it was estimated
that the probability of getting angiosarcoma at some time in the
worker's life per year of exposure is 0.0031. To arrive at this rate,
the analysis considered the distribution of exposure durations among
the 14 known occupational cases of angiosarcoma, as well as the time
distribution of man-years of exposure in the industry. Two other
important facts about vinyl chloride carcinogenesis resulted from this
analysis: 1) at some time in their lives about 7.5% of all highly-
exposed workers are expected to get liver angiosarcomas due to vinyl
chloride exposure with double this rate of primary cancer at all sites
combined. 2) Of all the cases of liver angiosarcoma which have this far
been produced by vinyl chloride, only 38% of them have been diagnoses
as of 1974.
After considering the error associated with the separate estimates
based on animal and human data, it is concluded that the two approaches
give results that are indistinguishable.
The projection to low doses was done using two separate mathematical
models of the animal data: 1) the linear model, which assumes that the
response rates are directly proportional to dose with no threshold;
2) the log-probit model, which assumes that the susceptibility of the
population is log-normally distributed with dose. The former model is
widely used for low-dose extrapolation in radiation carcinogenesis and
is generally considered to be conservative for chemical carcinogens, and
the latter is usually used for describing the dose-response relationship
in the dosage range of animal experiments.
It was found that the rate of initiation of liver angiosarcoma
among people living around VCM-PVC plants is expected to range from less
than one to ten cases of liver angiosarcoma per year of exposure to vinyl
chloride. The log-probit model gives predictions that are 0.1 to 0.01
times this rate. This wide range is an indication of the uncertainties
in extrapolation to low doses. The cases initiated by exposure this
year will not be diagnosed until 1991 or 1995. Vinyl chloride is also
expected to produce an equal number of primary cancers at other sites,
for a total of somewhere between less than one and twenty cases of cancer
per year of exposure among residents around plants. The number of these
effects is expected to be reduced in proportion to the reduction in the
ambient annual average vinyl chloride concentration, which is expected
-------
to be 5% of current levels after the proposed EPA regulations are
Implemented.
As part of this analysis a survey of the known cases of liver
angiosarcoma diagnosed in the last 10 years was initiated in order to
see whether any evidence could be found that living near VCM or PVC
plants results in higher rates than expected for the general U. S.
population. No conclusion can be drawn from the survey at its present
stage of completion. It is expected that if the true rate is as high
as 10 cases per year of exposure among residents near plants, which is
the highest rate our analysis would predict, higher than average rates
of case occurrence might not yet be observable in such a survey but
should become detectable in the next 5-10 years, due to the increase in
vinyl chloride production rates since 1955-1960.
-------
iv
Table of Contents
Executive Summary i
Quantitative Risk Assessment for Community
Exposure to Vinyl Chloride 1
Size of Exposed Population 1
Ambient Vinyl Chloride Concentrations 2
Health Effects Resulting from Exposure 3
Appendix A. Population and Exposure Estimates A-l
Population Estimates A-l
Ambient Concentration Estimates A-2
Overall Exposure Estimates A-4
Appendix B. Estimates of Effect Rates from
Animal Data B-l
The Animal Data on Cancer B-l
Extrapolation to Low Doses B-2
Extrapolation to Human Exposure B-4
Incidence of Non-Cancer Effects B-5
Overall Effect Rates B-5
Error Analysis B-6
Appendix C. Mathematical Treatment of
Dose-Response Data C-l
Equations for ML Estimators C-l
Asymptotic Variances C-2
Newton's Method for One Variable C-3
Newton's Method for Two Variables C-3
Appendix D. Use of Human Data to Estimate Risks D-l
Interspecies Comparison of Life Expectancy
and Angiosarcoma Latent Times D-l
Comparison of Angiosarcoma Incidence D-2
Alternative Approach to Calculating
Incidence Rate D-4
Comparison with Animal Data D-6
Ratio of All Cancers to Liver Angiosarcoma D-7
Frequency of Non-Cancer Effects Relative to
Liver Angiosarcoma D-8
Appendix E. Is Residence Near Vinyl Chloride Plants a
Risk Factor in Frequency of Deaths Due to Liver
Angiosarcoma? E-l
-------
V
Introduction E-l
Sources of Data E-l
Procedures of this Study E-3
Results of the Survey E-3
Discussion of Results E-5
-------
Quantitative Risk Assessment for
Community Exposure to Vinyl Chloride
Vinyl chloride is a known human carcinogen; it has produced liver
angiosarcoma, a very rare form of cancer, as well as other cancers and
non-cancer effects in occupationally exposed populations. It is also
known to be emitted into the atmosphere from plants which produce vinyl
chloride monomer (VCM plants) and plants with polymerize the monomer to
polyvinyl chloride (PVC plants). Although concentrations of vinyl
chloride in the ambient air are much less than those which caused cancer
in workers, it is generally considered prudent to assume that there is
no threshold for chemical carcinogens, so that any exposure involves
some risk. In conjunction with EPA consideration of rulemaking action
to regulate emissions of vinyl chloride from VCM and PVC plants, the
Administrator of EPA requested that an analysis be performed which
would estimate quantitatively the risk resulting from VC emissions and
assess the reliability of the estimates. This paper reports the results
of that analysis. The details are presented in the Appendices A
through E.
The analysis involves three steps which are discussed in turn below:
an estimate of the size of the exposed population, an estimate of the
concentrations of vinyl chloride to which they are exposed, and an
estimate of the number of liver angiosarcomas and other health effects
which would result from this exposure. Of these, the last estimate is
by far the most difficult to make. In addition an investigation was
made of the places of residence of all people known to die of liver
angiosarcoma in the last ten years in an attempt to detect clustering
around PVC and VC plants.
Size of exposed population
A study by the American Public Health Association (APHA), performed
under contract to EPA's Office of Toxic Substances, determined the
number of people living within various distances, up to 5 miles, from
each VCM or PVC plant. The study was based primarily on census tract
information. The validity of the methodology used was confirmed by a
more detailed analysis of the population living around a few plants,
performed by EPA's Office of Planning and Evaluation.
The total population living within 5 miles of all PVC and VCM
plants is shown in the following table:
Distance (mi)
Population
47,000
0 - 1/2
1/2 - 1
1 - 3
3 - 5
203,000
1 ,491 ,000
2.838,000
TOTAL
4,579,000
-------
2
Thus, a total of 4.6 million people live in the vicinity of these
plants. To calculate the average exposure, detailed population data
around each individual plant was weighted to reflect differences in
the type of plant (PVC or VCM), size of plant (average or large) and
meteorological conditions at the plant site.
The use of residence data involves some error, of course, since
people spend part of their time away from their homes and hence exposed
to varying levels of vinyl chloride. There does not seem to be any
practical way around this problem, short of a detailed study of travel
patterns of 4 million people in over 40 separate communities.
Ambient vinyl chloride concentrations
Annual average ambient concentrations of vinyl chloride were
calculated by standard diffusion modeling techniques. Two independent
studies were made, one by EPA's Office of Air Quality Planning and
Standards (OAQPS) and one by Teknekron, Inc. The agreement among the
two studies was good, with differences generally less than 25%. It
was decided to use the Teknekron results in the actual calculations
since they included data on variations in meteorological conditions
from location to location.
For an average uncontrolled PVC or VCM plantin an area with average
meteorological conditions, the annual average concentration of vinyl
chloride in each annulus around the plant is shown in the following
table:
Distance (mi) Vinyl chloride concentration (ppb)
PVC plant VCM plant
0 - 1/2 323 113
1/2 - 1 57 20
1-3 15 5.2
3-5 5.7 2.0
It can be seen that concentrations around VC plants are significantly
smaller than around PVC plants. This fact combined with the much
smaller population living near VC plants, implies that by far the
greatest part of the public health risk is from emissions from PVC
plants.
In calculating total population exposure, it is necessary to take
account of factors that lead to variations from the average. Information
from OAQPS and from the APHA study was used to determine areas where more
than one plant was located, and OAQPS characterized the size of each
plant as "average" or "large." The Teknekron study was used to categorize
the meteorological and topographic conditions at each location. Details
are given in Appendix A.
-------
3
The net result of these calculations is that the average exposure
faced by a person chosen at random from the 4.6 million people living
within 5 miles of plants is 17 ppb.
Unfortunately, it has not been possible to make a systematic com-
parison of the diffusion modeling results with data obtained from actual
monitoring, although they appear generally consistent. It is therefore
difficult to estimate the uncertainty of these estimates. Lacking
anything better, we can take the difference between the two diffusion
modeling efforts of up to 25% as an estimate of that uncertainty.
Health effects resulting from exposure
What are the results of exposing 4.6 million people to an average
of 17 ppb of vinyl chloride? The first major decision to face in
answering this question is to arrive at some combination of two basic
approaches. One approach is to rely largely on human data (which exists
for vinyl chloride but not for many other chemicals of concern to EPA):
the secons is to make projections from animal experiments. Both involve
difficulties. Use of human data eliminates the uncertainties that
result because we do not know the differences in response between the
test animals and humans. On the other hand, with the data on human
(occupational) exposure, it is necessary to guess at exposure levels
over the past 30 years and approximate the total number of workers
involved and the number of cancers caused by past exposures for which
symptoms may not appear until many years in the future. By using animal
data we can avoid these problems, but only at the price of uncertainty
in the relevance of animal experiments to human exposures. The approach
taken 1n this analysis is to use animal data to predict the probability
of human liver angiosarcoma, and then use the human data to the greatest
extent possible to interpret those predictions.
A second major decision that must be made is how to project the
results observed at high doses in animal experiments (and in the occupa-
tional exposures) to the much lower doses encountered in the environment.
Two alternative assumptions are frequently made in the scientific litera-
ture. The first is a straight-line projection to zero dose, assuming no
threshold (the "linear model"). This is also referred to as the "one hit"
model, since it would follow logically from the assumption that each
minute increment of exposure to a carcinogen has the same independent
probability of causing a cancer, regardless of the dose level. This
assumption is generally accepted as prudent in radiation carcinogenesis.
For chemical carcinogenesis, the model is usually considered to provide
an upper limit to the level of effects likely at extremely low doses,
because the existence of detoxification mechanisms would render small
doses less effective in causing cancer and would therefore result in a
threshold of at least fewer effects.
The second commonly used projection method is based on the assump-
tion that the observed changes in response with dose are the result
-------
4
of variations of susceptibility in the population, which is assumed
to be log-normally distributed with dose. For convenience, we refer
to this as the "log-probit" model because it forms a straight line when
the logarithm of the dose is plotted against the proportion of responses
expressed in probability units (probits). The log-probit model is
closely related to the Mantel-Bryan procedure.
In this analysis, both models are used. For technical reasons,
the log-probit model is difficult to apply to this case. Therefore,
the basic calculations were done using the linear model, but a sen-
sitivity analysis was done to show how the results would change under
the log-probit assumption.
A third decision that must be made is how to predict human
incidence rates from animal data. Again, there is little hard data
to provide guidance. The assumption used here is that a lifetime
exposure of humans to a given concentration of vinyl chloride would
produce the same number of effects as a lifetime exposure of rats. Thus,
the one-year exposure in the animal experiments would be equivalent to
about 30 years of exposure for humans.
A fourth decision to make is how to use the available human data
on liver angiosarcoma cases among highly-exposed workers to calculate
the probability per year of exposure that cases will eventually develop
in people. This calculation is needed for comparisons with the animal
model. There are three aspects to this problem: 1) to find in the
literature a realistic estimate for the fraction of highly-exposed
workers who have contracted liver angiosarcoma at some time in their
lives; 2) to account for the fact that the currently-observed rates
underestimate the actual incidence because they do not include workers
who have been exposed more recently than 15 to 20 years ago; and 3) to
account for the fact that people can die from other causes before a
latent case of liver angiosarcoma becomes manifest.
These issues were treated as follows: Of the four occupational
epidemiology studies from which it is possible to estimate an
incidence rate, the two with the smallest number of subjects and the
best separation of highly-exposed workers from the group of all workers
had the largest incidence of angiosarcoma. This incidence was assumed
to be valid for all highly-exposed workers. The latency time for liver
angiosarcoma and the growth in the number of person-years of exposure
since 1940 are two factors which affect the number of cases we have
observed through 1974. These factors, along with the distribution of
exposure durations for people who have been diagnosed with liver
angiosarcoma, were included in an analysis presented in Appendix D.
The result of the analysis is an estimate of the probability per year of
exposure that a person will get angiosarcoma sometime in his life. The
remaining problem of multiple risks competing for mortality was not
treated because of its complexity.
-------
5
A fifth decision which had to be made was how to qualitatively
describe the other effects of vinyl chloride exposure besides liver
angiosarcoma. This problem was handled by estimating from the
literature ratios of the number of people with other cancers and the
number of people with liver damage compared to the number with angio-
sarcoma. As an index of liver damage, the bromsulphalein (BSP) test
is used because it, among all liver function tests that have been used,
correlates best with vinyl chloride exposure and because an abnormal
BSP test indicates that severe damage has occurred in the liver, either
because the liver cells are not able to assimilate the intravenously
injected BSP dye from the blood and excrete it into the bile passages,
or that the bile passages are no longer structurally intact enough to
carry the dye out of the liver.
The results of these five aspects of the problem are presented
below in reverse order. The approximate ratio of severe liver damage
cases to liver angiosarcoma cases is about 30, the result being con-
sistent for two independent occupational studies. It was also found
that about twice as many cases of cancer of all sites are caused by
vinyl chloride as cases of liver angiosarcoma alone. The animal
experiments have shown approximately the same ratio of all cancers to
liver angiosarcoma, after background incidence is taken into account.
In calculating the probability per year of exposure that a highly-
exposed worker will get angiosarcoma some time in his life, we found
that the fraction of highly-exposed workers who have been currently
diagnosed is 0.02; they have been exposed for an average of 17 years
before diagnoses. The analysis,which required assumptions about
the time distribution of man-years of exposure and the distribution of
exposure durations preceding diagnosis, showed that only 38% of the
highly-exposed workers who are expected to get angiosarcoma some time
in their lives have been already diagnosed. Therefore the probability
that one of these people will get angiosarcoma some time in their lives
is 0.02/(17 x 0.38) = 0.0031 per year of exposure. In Appendix D the
calculation is explained in greater detail.
i
The 17-year average concentration to which these workers were
exposed was estimated to be 350 ppm on the basis of one study. Only
one company has reported measurements of vinyl chloride for the jobs in
their plant. These measurements, started in 1950, showed the highest
exposure jobs ranged from 120 to 385 ppm before 1960, when the
exposures were reduced because of suspected toxicological problems with
vinyl chloride. In estimating the average, it was assumed that the
other factories, most of which probably did not monitor the concentra-
tion of vinyl chloride, were less concerned about industrial hygeine and
therefore took fewer pains to keep the levels low.
In predicting the human angiosarcoma rate from the animal dose-
response data, it was projected, from the linear model, that exposure
to vinyl chloride would cause 0.071 cases of liver angiosarcoma and
-------
6
0.15 cases of all types of cancer per million people per year per ppb of
continuous exposure. Details of these calculations are given in
Appendix B. Converting to a 7-hour per day, 5 days per week work
schedule of exposure to 350 ppm, the model predicts an incidence rate
of 0.0052 per person-year exposure. It is shown in Appendix D that
this rate is numerically indistinguishable from the rate of 0.0031
calculated from the human data, considering the known quantifiable
errors of estimating the parameters of the animal and human data. It
can be concluded that the slope of the linear animal dose-response
relationship for angiosarcomas is consistent with the human data.
The extrapolation of the animal dose-response relationship to a
concentration of 17 ppb (the average concentration around the uncon-
trolled plants) yields the following predicted number of cases in the
4.6 million people living within 5 miles of the plants. For details
see Appendix B.
Cases per year of exposure
Type of effect Linear model Log-probit model
All cancer T1 0.1 - 1.0
Liver angiosarcoma 5.5 0.05-0.5
This is the expected number of cases projected to be caused per
year at current levels of emissions; the people exposed now will not
be diagnosed for another 15-20 years. Similarly, any cases observed
now would have been caused 15-20 years ago (if they were in fact caused
by vinyl chloride) when production was about 10% of current levels.
In order to arrive at a final estimate of the number of people
adversely affected by vinyl chloride emissions, the important results
of this analysis to consider are as follows: 1) the number of cancers
at all sites caused by vinyl chloride is twice the number of liver
angiosarcomas; 2) the number of people with severe liver damage is 30
times the number of liver angiosarcomas; 3) the animal model predicts
that the number of liver angiosarcomas in the population around plants
is 5.5 cases per year of exposure; 4) the number of cases calculated
from the human data is 60% of the number predicted from the animal
model; 5) the use of a log-probit model for extrapolation to low doses
gives predictions of 0.1 to 0.01 times the number predicted by the
linear model; 6) the error in the estimate of 5.5 cases per year ranges
from + 55% to -10%. This error includes statistical uncertainty in
estimating the dose-response, uncertainty in ambient concentration
estimates, and errors resulting from not considering exposures beyond
5 miles or decomposition of vinyl chloride in the atmosphere. It is
not symmetrical because it includes possible effects beyond 5 miles from
the plants, which were not explicitly considered in the analysis. It does
not account for our uncertainty about the appropriateness of using a
-------
7
linear model extrapolated to zero dose; or of extrapolation from animal
data; 7) the quantifiable error in the rate calculated from the human
data is about + 67%. This includes uncertainties in the 17-year average
dose received By the workers, uncertainty in the number of hours per day
of actual high exposure, and uncertainty in the fraction of highly-exposed
workers who have been diagnosed with liver angiosarcoma. Other errors
can not be quantified, and are discussed in Appendix D.
When all these uncertainties are considered, our judgement is that
the number of liver angiosarcoma cases produced per year of exposure in
people residing near vinyl chloride plants is somewhere between less
than one and 10 cases. The cases produced by this year's exposure will
not be diagnosed until 15 to 20 years from now. If the EPA regulations
are implemented, the number of cases is expected to be reduced in
proportion to the reduction in the ambient annual average concentration,
which is expected to be 5% of the present level.
The vinyl chloride exposure around plants is also producing
somewhere between less than 1 and 10 cases of primary cancer at other
sites, mainly lung, brain and bone. Assuming no threshold for liver
damage somewhere between less than 1 and 300 cases of serious liver damage
would be predicted. The number of liver damage cases is likely to be
less than this because the assumption of no threshold is likely to be
wrong.
In order to find out whether people living near VC-PVC plants have,
as of 1974, had higher rates of liver angiosarcoma diagnosis than the
overall U, S. population, a search of the residence records of all known
liver angiosarcoma cases in the last 10 years was performed. Out of 176
cases where residence at time of death was known, 3 people lived within
5 miles of a plant. Unfortunately, the diagnosis of these cases has not
yet been confirmed by the National Cancer Institute. In addition one
infant whose parents lived within one mile of a plant died of a relatively
common liver tumor. It was shown in Appendix E that this rate of occur-
rence is not higher than the national average. However, the survey is
too incomplete to draw any conclusions at the current time.
Considering the results of the foregoing analysis, one would only
now expect to be seeing some evidence of vinyl chloride exposure. If
the highest rate in our range were actually occurring, 10 cases of liver
angiosarcoma per year of exposure are being produced; 15-20 years ago
when the vinyl chloride production 3as 10% of current levels, one case
would be expected per year of exposure (with constant population). This
is to be compared to a background rate of 0.6 cases per year that are
occurring now.
The survey of liver angiosarcoma cases would probably detect the
existence of 10 cases over the 10 year period. Since this was not
observed we can conclude that the real incidence is not significantly
greater than the predicted upper limit of 10 cases initiated per year
-------
8
of exposure unless migration of people in and out of the regions around
plants has been excessive. If the lower rates in the range of the above
analysis were to be true, increased incidence of angiosarcoma would not
be observable.
-------
Appendix A. Population and Exposure Estimates
This appendix describes the approach taken in estimating the
population at risk and the levels of exposure. The task is made easier
by the use of a linear dose-response model to project health effects at
very low exposures. This model implies that, for example, a given
number of people exposed to emissions from two plants, each at the
same distance from the people, would suffer the same number of health
effects as twice as many people at the same distance from one plant.
More formally, with a linear dose-response relationship, the
number of health effects is proportional to the sum, over all plants,
distances, etc. , of population times dose. This discussion shows
how this quantity was estimated.
Population Estimates
The population estimates used in this analysis are all based on a
study performed by the American Public Health Association under
contract to the Office of Toxic Substances (APHA, Population Residing
Near Plants Producing Vinyl Chloride, Aug. 1975, hereinafter cited as
"APHA Study").
The APHA Study is based on the 1970 Census. For each PVC or
VCM plant (or group of plants located close together), APHA determined
the number of people residing within 0-1/2 mi, 1/2-1 mi, 1-3 mi and
3-5 mi. Further breakdowns by age and sex and by direction from the
plant were also determined; however, as is discussed below, these
detailed breakdowns were not used in this analysis.
For those plants located in Standard Metropolitan Statistical Areas
(SMSAs), where Census Tract data is available, APHA asked a local
respondent to allocate each tract to one of the distance ranges. Census
tracts were not split. Independently, the Office of Planning and
Evaluation, EPA, made a similar analysis for several selected plants
in which Census tracts were split and one for which city block data was
used. A comparison with the APHA data for those plants indicates that
no significant errors were introduced by the APHA procedure. For
untracted areas, population was assumed to be uniformly distributed
over the area. For a more detailed description of the methodology used
to estimate population, the reader is referred to the APHA report.
The total population residing at the indicated distances from PVC
or VCM plants is shown in the following table:
Distance (mi) Population
0-1/2 47,000
1/2-1 203, 000
1-3 1,491, 000
3-5 2,838, 000
Total
4, 579, 000
-------
A-2
Of course, some of these individuals are exposed to VC from more than
one plant or are exposed to emissions from plants that are larger than
average. These variations are taken into account in calculating exposure
estimates, as described below.
There are several sources of error which result from the methodo-
logy used here. First, by estimating the exposed population from
Census statistics on place of residence, we do not take into account the
daily mobility of people, some of whom travel into areas of higher VC
concentrations than their residences while others do the opposite.
There does not seem to be any way of estimating the net effect of such
travel or even whether it increases or decreases overall exposure, short
of an analysis of commuting and travel patterns in the 42 communities
included in this study. In general, it might be noted, however, these
plants do not appear to be located in large central business districts to
which large number of non-factory workers commute. Given the much
greater uncertainties that are inherent m the estimation of dose-response
relationships, it does not appear to be worth the effort to do such
anaylses.
Second, the overall U. S. population has grown about 5% since the
1970 Census, but it is not known whether the areas covered in this study
have grown similarly. The error here again seems small compared with
other uncertainties.
Third, people living further than 5 mi from plants are not included.
The effects of this error are considered below in the context of the
overall calculation.
Fourth, even though information was available on the distribution of
exposed populations by age, sex, and direction from the plant, this data
was not used. The data on age and sex was not used because there is no
information on how susceptibility to the effects of exposure to VC would
vary with these factors. The data on distributionn of population by
direction from the plants could have been used, together with data on the
distribution of the ambient VC concentration by direction, to develop a
more accurate estimate of the overall population exposure to VC. An
examination of several communities with large exposed populations
suggested, however, that the populations were either distributed in many
directions or did not appear to be correlated with the angular distribution
of VC concentration. Since the more detailed analysis would require 16
times as much calculation, it was judged not be worth the effort, parti-
cularly given the much greater uncertainties in other parts of the analysis.
Ambient Concentration Estimates
It was decided early that the measure of exposure to be used is the
long-term (annual) average ambient concentration. At one level, this
assumption is a consequence of the linear (single shot) dose-response
relationship. But regardless of the specific form ot the dose-response
relationship, it appears reasonable to assume that the probability that
-------
A-3
any given individual suffers an adverse health effect would depend on the
long-term integrated exposure of the individual, independently of the
functional relationship between the probability and the exposure level.
Indpendent of the actual truth of this assumption, however, is the absence
of any data on responses to peak exposures, either from the occupational
experience or animal experiments. Thus, there is no alternative to this
assumption if a quantitative projection of effects is to be made.
The estimates of average annual concentrations were derived from
diffusion modeling, using standard techniques. Two independent modeling
efforts were made. The first was performed by the Monitoring and Data
Analysis Division of EPA's Office of Air Quality Planning and Standards
in Research Triangle Park, N. C. This analysis took one representative
set of meteorological conditions and devoted considerable effort to
investigating the effects of different plant source configurations for con-
trolled and uncontrolled PVC and VCM plants. In general, for average
annual concentrations, the results for uncontrolled PVC plants were
quite similar for the plant configurations analyzed, as were those for
uncontrolled VCM plants (although, of course, emissions from PVC plants
are very different than those from VCM plants). The results for typical
PVC and VCM plants are shown in Figs. A-l through A-4.
The second independent effort was performed by Teknekron, Inc.
under the contract with EPA's Office of Planning and Evaluation. In this
analysis, plant configurations were fixed (except for size and the differ-
ence in pattern of emissions between PVC and VCM plants), but an effort
was made to estimate the distribution of meteorological conditions and,
to some extent, topography for each site.
The results of the two sets of calculations were extremely close.
The following table shows the average annual concentration of VC emitted
from a typical PVC plant as a function of distance (that is, averaged
radially through all directions), as determined by the two studies.
Vinyl Chloride Concentration (ppb)
Distance (mi) OAQPi? Teknekron
0. 25
301
323
0.75
76
57
1. 00
--
37
1. 50
26
2. 00
15
3. 00
10
8
4. 00
--
5. 7
4. 50
5. 6
5. 00
4. 0
A dash (--) in the table indicates that a value was not determined for
precisely that range. The results appear generally to be within 25% of
each other, with the OAQPS results lower than the Teknekron results at
very close ranges and higher at longer ranges.
-------
A-4
In general, the closeness of the two sets of independent results
serves as a confirmation of both. Of course, since both use similar
methodologies, it is possible that they share some of the same errors.
It has not proved possible to compare these modeling results with actual
monitoring data in a systematic way, and this is an important shortcoming
of the analysis as it stands at present. Summaries of the ambient measure-
ments have been examined, however, and they appear to be generally con-
sistent with the modeling results.
For the actual calculations, it was decided to use the Teknekron
results, since these showed how ambient concentrations would vary with
meteorological conditions (although, in retrospect, it turned out that these
variations made little difference in the final result). The sites listed in
the Teknekron report divided naturally into four categories, which we
designate as Low (L), Average (A), High (H), and Very High (VH), based
on the ambient concentration projected to result from the same level of
emissions. (Variation among sites due to the number and size of the
plants are factored into the analysis separately and are discussed below).
Table A-l shows the category to which each location is assigned. The
relative ambient concentrations for the four categories are:
OAQPS provided information which categorized each plant by size as
"Average" or "Large. " The typical "Average" VCM plant had a production
of 700 million lb/yr, and the typical "Large VCM plant 1300 million lb/yr.
For PVC plants, the typical "Average" plant used 150 million lb/yr and
the typical "Large" plant 350 million lb/yr. It was assumed that large
plants would have proportionately greater emissions. It is estimated that
uncontrolled PVC plants emit about 4% of the vinyl chloride processed and
VCM plants, about 0.3%. Table A-l also shows, for each location, the
number and size categories of plants located there. The letter "L"
indicates a "Large" plant; thus, for example, the designation "2" means
two "Average" plants and "2 & 2L" means two "Average" and two "Large"
plants.
Overall Exposure Estimates
Table A-l also shows the populations residing at the indicated
distances from individual plants. To determine the equivalent exposure
(compared to exposure from a single average size plant at a location with
average meteorological conditions), each of the population figures must
be multiplied by two factors, one reflecting meteorological conditions at
that location and one reflecting the number and size of plants at the loca-
tion. Thus, for example, in Texas City, Texas, each individual is
exposed to emissions from two average size plants and one large plant
(equivalent to 2. 3 average plants) but under "Low" meteorological condi-
tions (leading to 0. 56 of the concentration that would be expected under
average conditions). Thus, these individuals are exposed to the equiva-
lent of 0. 56 x 4. 3 = 2. 4 average plants' emissions under average
Low
Average
High
Very High
0. 56
1. 00
1. 55
2. 55
-------
A-5
conditions. Similar calculations were performed for each location in
Table A-l and the results summed to obtain the values shown as "Weighted
total population" in Table A-2.
The weighted total population figures are then multiplied by the aver-
age annual concentration of vinyl chloride, as derived from the diffusion
modeling. The sum of these figures is the total population exposure.
These results are shown in Table A-2. The total exposure is estimated
to be 76.4 million people x ppb. Thus, the total exposure of the U.S.
population to vinyl chloride due to emissions from these plants is
equiwalent to 76. 4 million people exposed to 1 ppb, or 7. 64 million people
exposed to 10 ppb, etc. Probably the simplest way to understand this
figure is in terms of the average exposure of those who are exposed.
Since there are a total of 4. 6 million people exposed within 5 mi of plants,
the average exposure of these people is 76.4/4.6 or 17 ppb.
We can also use these figures to gain some insight into the distribu-
tion of the aggregate exposure (and hence of the resulting adverse health
effects). We can see from Table A-2 that only about 3% of the exposure
is a result of emissions from VCM plans, with the remaining 97% coming
from PVC plants. It also appears that the bulk of the health effects will
be occurring from 1 to 5 miles from the plants, rather than within one
mile, since much larger numbers of people live at the greater distances.
-------
Kikxnatan
0 1 2 3 A
I .....li.i.t ...«l
Figure A-3. Estimated annual ooncantrstions
(ppm) of VCM from attrage
sin VCM plant.
-------
.0025
.005
PLANT
.0025
Kilometers
0 3 6
J i I I I
Figure A-4. Estimated annual concentrations
(ppm) of VCM from average size
VCM plant.
-------
TABLE A-l Population Exposed by Location
No. & Size Meteorological Population vs. Distance (mi)
Location
of Plants
Conditions
0-1/2
1/2-1
1-3
3-5
(a) PVC Plants
Carson City, CA
2
A
—
16,541
259, 309
405,461
Saugus, CA
1
H
10, 935
--
12, 029
Delaware City, DE
1&1L
A
--
6,970
9, 218
Pace, FL
1
A
--
7, 071
Henry, IL
IL
A
37
146
1, 161
2, 013
Illiopolis, IL
1
L
30
117
922
1, 597
Calvert City, KY
1
A
102
398
3, 134
5,436
Louisville, KY
IL
A
6, 315
111, 146
146, 000
Plaquemine, LA
1
A
177
673
5, 328
9, 237
Baton Rouge, LA
1
A
--
25, 615
54, 193
Perryville, MD
IL
H
2, 091
11, 091
9,791
Fitchburg, MA
3
A
34,395
67,767
114,489
Springfield, MA
1
A
8, 258
40, 671
108,966
Midland, MI
1
L
492
1,899
15, 045
26,067
Aberdeen, MS
IL
VH
114
442
3, 504
6, 073
Burlington, NJ
2
L
8, 159
122,419
230, 361
Pedrickton, NJ
1
A
--
--
9, 113
98,992
^assaic, NJ
1
L
22, 512
29,458
194,601
269,259
. Kearny, NJ
1
L
18,656
214,594
696,928
Flemington, NJ
1
A
81
306
2, 394
4, 153
Hicksville, NY
1
L
5, 900
20,702
143, 760
215, 230
Williamsville, NY
1
L
11,522
12,552
51,872
113,422
Ashtabula, OH
1
A
434
1,671
13,239
22,937
Huron, OH
1
L
98
380
2, 988
5, 175
Avon Lake, OH
IL
L
827
5, 511
16, 398
8, 022
Painesville, OH1
1&1L
A
--
21,550
16,434
Oklahoma City, OK
IL
L
--
--
25,526
73,581
Pottstown, PA
IL
H
--
2, 447
36,121
30, 382
Deer Park, TX
IL
H
12, 820
32, 151
86, 508
Freeport, TX
1&1L
L
4, 746
15, 879
Texas City, TX
2&1L
L
3, 440
19,460
34, 751
Pt. Pleasant, WV
1
H
263
1, 007
7, 984
13,836
S. Charleston, WV
1
VH
5, 698
40, 685
53,432
(b) V CM Plants
Geismar, LA
3
A
62
233
1, 852
3, 211
Baton Rouge, LA
1
A
--
--
25, 615
54,193
Lake Charles, LA
2
A
11, 601
44,100
Plaquemine, LA
1
A
177
673
5, 328
9, 237
Norco, LA
2
A
215
823
6, 525
11,306
Deer Park, TX
2
H
12,820
32, 151
86,508
"reeport, TX
1
L
--
4, 746
15, 879
Jalvert City, KY
IL
A
102
398
3, 134
5, 436
Guayamilla, PR
1
A
202
780
6, 173
10, 692
-------
FOOTNOTES:
a/ Numbers followed by "L" indicate "Large" plants as defined
in the text. Other numbers indicate "Average" size plants,
b/ Ratio of ambient concentration to emissions is characterized
as Low
-------
TABLE A-2 Aggregate VC Exposure
(a) PVC Plants
Distance From Plant (mi)
TFT72 TTtt r^3 3^5 Total
Weighted total population 34,000 228,000 2, 013, 000 3, 439,000
VC concentration (ppb) 323 57 15 5.7
Total exposure 11.0 13.0 30.2 19. 6 73.8
(million population x ppb)
(b) VCM Plants
Weighted total population 1, 200 37, 000 169, 000 434, 000
VC concentration (ppb) 113 20 5.2 2.0
Total exposure 0. 1 0. 7 0.9 0.9 2.6
(million population x ppb)
Total for all plants
76.4
-------
Appendix B. Estimates of Effect Rates from Animal Data
In analyzing the effects of vinyl chloride, we are fortunate to
have at least semi-quantitative data for both human and animal expo-
sures, Choices had to be made concerning how to use this data.
Vinyl chloride is known to cause both cancer and non-cancer effects
in humans. With respect to cancer effects, we have animal experi-
ments in which rats were exposed to known concentrations for known
periods of time. We also have the human occupational experience, in
which increased incidence of liver angiosarcomas and other cancers
were observed in exposed workers. Ideally, we should use human data
throughout and avoid the problem of extrapolating from an animal model
to human beings. Unfortunately, we do not have good data on the level
of exposure of workers. In arriving at a specific number to write down
and use in the calculations, it would be necessary to guess at these fac-
tors. We decided, therefore, to derive a dose-response relationship
from the animal data, for cancer effects, but to use the human data to
the greatest extent possible to confirm or deny the validity of the relation-
ship. These checks are described in Appendix D. Since both the human
and the animal data involve exposure at high levels, the difficulty in extra-
polation from high to low exposure levels would be the same in either case.
With respect to effects other than cancer, the animal data is too
fragmentary to estimate quantitatively the ratio between cancer and
non-cancer effects caused by exposure to vinyl chloride. However,
the human data does permit an estimate of this ratio, and this estimate
can then be used together with the dose-response estimate for cancer
to place some limits on the magnitude of the non-cancer effects.
The Animal Data on Cancer
The basic experiment is Maltoni's experiment BT1. Sprague-
Dawley rats were exposed by inhalation to a specified concentration
of vinyl chloride for 4 hr/day, 5 days/week for 52 weeks. After the
52 weeks, the animals were not exposed to vinyl chloride but were
observed for the remainder of their lifetimes. The results of the
experiment were as follows:
VC Concentration Liver
(ppm)
All Tumors
10, 000
6, 000
2, 500
500
250
50
0
9/61 (15%)
13/60 (22%)
13/59 (22%)
7/59 (12%)
4/59 ( 7%)
1/59 ( 2%)
0/58 ( 0%)
38/61 (62%)
31/60 (52%)
32/59 (54%)
22/59 (37%)
16/59 (27%)
10/59 (17%)
6/58 (10%)
-------
B-2
Two conclusions can be drawn immediately from these results.
First, the rate of occurrence of both liver angiosarcomas and total
cancers increases much more slowly with dose above 500 ppm than
below this concentration. The cause of this phenomenon is not under-
stood, but, whatever it may be, it does not appear to be relevant to
extrapolation to lower doses in the parts-per-billion range. Hence,
all the curve-fitting reported here is done only for the data for 500
ppm and below.
Second, the control group had no liver angiosarcomas, but about
10% of them did have other types of cancers, as is to be expected.
Therefore, in estimating the dose-response relationship for total
cancers, it will be necessary to use a method which allows for a
positive response at zero dose.
Extrapolation to Low Doses
It is widely understood in the scientific community that no firm
logical basis can exist for extrapolating data to levels many orders
of magnitude beyond the range of experimental observation. In
spite of this, two alternative assumptions are frequently made in the
literature. The first of these is that the response is linear m dose,
e. g., half as many effects are produced at half the dose. This is
also referred to as the "one-hit" model, since it would follow
logically from the assumption that each minute increment of expo-
sure to a carcinogen has the same independent probability of causing
a cancer regardless of the dose level. This assumption is generally
accepted as prudent in radiation carcinogenesis. For carcinogenesis
in general, this model is usually considered to provide an upper limit
to the level of effects likely at extremely low doses, because the
possible existence of detoxification mechanisms might render small
doses ineffective in contributing to cancer effects and thus result in
a threshold below which no effects would occur. The second assump-
tion frequently made is that the response is linear when the logarithm
of the dose is plotted against the proportion of responses expressed
in a probit scale (i. e., the value of the argument of the cumulative
normalized Gaussian distribution function corresponding to that
proportion); for convenience, we refer to this as the "log-probit"
model. This model is derived from the assumption that individuals
in the population have a statistical distribution of susceptibility to
cancer effects which is normally distributed with log dose.
The linear model is much easier to apply, since the number of
effects depend only on the number of individuals exposed and their
average dose. The log-probit model, on the other hand, requires
data as to the distribution of doses in the population and is therefore
more likely to be influenced by anomalies in the methods for estimat-
ing population exposure. For this reason and because of the conser-
vativeness of the linear model, the calculations were made using the
linear model, but a sensitivity analysis was performed to show how
the results would vary were the log-probit model used.
-------
B-3
Straight lines were fitted to Maltoni's data using a maximum -
likelihood method. Essentially, this method assumes that the prob-
abilities of cancer at the different doses were related by a linear func-
tion and detemines the values of the slope and intercept that would
make the likelihood of the observed results as high as possible.
Mathematical details are provided m Appendix C.
For the data on liver angiosarcoma, the line was constrained
to go through the origin. (Without this constraint, a slightly nega-
tive intercept was obtained. ) With the constraint, the following
relationship is obtained:
where P is the probability of a liver angiosarcoma and cl is the dose
in ppm. For all tumors, the corresponding line is —
Here, the intercept is the best estimate from all data at 500 ppm and
below of the cancer rate m the absence of vinyl chloride; the slope is
the best estimate of the rate at which tumors are caused by exposure
to vinyl chloride. Thus, inn the animals, vinyl chloride appears to
cause about one other cancer for each liver angiosarcoma. Estimates
of the total number of cancers caused by vinyl chloride are based on the
slope of this line, not on the intercept.
The standard deviations of the estimates of the slope were found
to be 6. 9 x 10 for liver angiosarcoma and 1.1 x 10~4 for aii tumors.
(These estimates are derived from large-sample theory). It cannot
be emphasized too much that the standard deviations reflect only that
uncertainty in the estimates that is due to limited sample size; they
are derived from the linear model and cannot reflect the much greater
uncertainty due to our ignorance of whether the linear model is correct.
They also do not reflect the effects of the part-time exposure used in the
experiments or, most importantly, of the difference between human
beings and Sprague-Dawley rats; these are discussed below.
The goodness of fit is shown in the following table, which compares
the actual data obtained in the experiments with the values expected
from the above equations.
P = 2. 53x10"4d
P = 0. 123 + 5. 26x10"4d
Dose (ppm) Actual
Liver Angiosarcoma
Actual Expected
Actual Expected
All Tumors
0
0
1
4
7
0. 00
0. 75
3. 74
7.47
6
10
16
22
7. 16
8.84
15. 05
22. 81
50
250
500
In order to test the sensitivity of these projections to the assumed
linear form of the dose-response relationship, the data for liver angio-
sarcoma was also fitted to a log-probit equation. Using an unweighted
-------
B-4
least-squares fit to the data for doses of 500 ppm or less, the
following equation is obtained:
probit = -3. 71 + 0. 932 log dose
where log indicates common (base 10) logarithm and dose is in ppm.
The fit is extremely close: the maximum difference between predicted
and actual values is 0. 02 probits.
Both the linear and log-probit fitted curves are plotted in Fig. B-l
on a log-probit scale. It can be seen from this graph that, in the
range of concentrations of interest (say, 5 to 300 ppb), the log-probit
projection yields risks of one to three orders of magnitude lower
than the linear projection. Since the result of a complete log-probit
analysis of human effects due to exposure to vinyl chloride in the
ambient air near plants would tend to be dominated by the high concen-
trations close to the plants (unlike the linear analysis), the log-probit
analysis would yield final results one to two orders of magnitude lower
than the linear model.
Extrapolation to Human Exposure
The equations derived above can be used to project what the
response of rats would be in similar experiments at much lower
doses. To apply these results to human community exposures, it is
necessary to modify the results to reflect 24-hour exposure and, most
importantly, the difference between humans and rats. The latter is
particularly problematical. The remainder of this Appendix will be
based on animal data. Human data will be discussed in Appendix D.
In the experiments, the animals were exosed to vinyl chloride
for 4 hr/day, 5 days/week. In the community, the exposure would be
continual for 24 hr/day, 7 days/week. This effect can be corrected
for by dividing the continuous dose received by people by (7x24)/
(5x4) = 8. 4 and using the same linear dose-response function to cal-
culate the risk. This is mathematically equivalent to multiplying the
slope by the same amount.
Adjusting for the difference between humans and rats is much more
difficult. As in the extrapolation to low doses, very little evidence is
available. Furthermore, controlled experiments involving human
carcinogenesis are unlikely to be carried out. In spite of the lack of
evidence, the consensus of the scientific community appears to be
that, if an assumption must be made, it should be assumed that the
same number of cancers would be induced in humans over their life-
times as are induced in rats over their lifetimes. With this assump-
tion, the one-year exposure of the experimental is expected to cause
the same probability of cancer (i. e., the same incidence rate) as
about 30 years of exposure to humans at the same concentration.
Thus, to determine the annual rates of cancer induction in humans,
-------
B-5
the rates derived above should be divided by 30.
Making these adjustments and also converting the units to cases
per million population per ppb per year, we obtain a rate of 0. 071 liver
angiosarcomas and 0.147 total cancers (in excess of background rates)
per mllion population per ppb vinyl chloride per year. The standard
deviations of the estimates, which, again, reflect only the statistical
uncertainty due to limited sample sizes, are 0. 019 and 0. 031,
respectively.
Incidence of Non-cancer Effects
Liver damage has been a frequent observation in animals exposed to
vinyl chloride. The lowest dose in chronic experiments (times longer
than 3 months) at which non-cancerous liver effects are observed is 100
ppm (Torkelson, et al., 1961). The effect observed was liver enlarge-
ment which cannot be called damage since the microscopic structure was
normal, but it is an indication that the liver is affected in some way by
vinyl chloride. Of all reports on animals the lowest dose causing liver
cellular degeneration is 500 ppm (7 hrs/day, 5 days/week, for 4-5
months). A Russian report of bone resorption, cardiovascular impair-
ment and neurological changes in the hypothalmus of rats and rabbits at
concentrations of 12-16 ppm for 6 months (Basalaev, 1972) has also
appeared. It is unfortunate that we currently have no published reports
of pre-cancerous liver damage occurring in the same experimental
group of animals in which cancer later develops. Therefore there is
no way to estimate the incidence ratio of non-cancer to liver
angiosarcoma effects in animals.
The human data on non-cancer effects is discussed m Appendix D.
Overall Effect Rates
In Appendix A, it was estimated that the 4. 6 million people living
within 5 mi of PVC or VCM plants would be exposed, on the average,
to 17 ppb of vinyl chloride. Combining this information with the dose-
response estimates derived above, we can calculate that, for example,
vinyl chloride emisssions in the absence of control would cause 4. 6 x
17 x 0. 071 = 5. 5 cases of liver angiosarcoma per year in the U. S. ,
using the linear model. Similar calculations for all cancer yield the
following results:
Type of Effect
Cases Caused per Year
Cancer
Liver angiosarcoma
All cancer
11
5 1/2
It should be realized that cancers caused by vinyl chloride have
about a 20-year latent period. Thus, we would not expect to observe
these numbers of cancers occurring now; rather, these are the numbers
-------
B-6
that are estimated to be initiated now but which will produce symptoms
perhaps 20 years from now if emissions remain unchanged from current
uncontrolled levels. Similarly, we would expect that cancers caused by
vinyl chloride that are showing symptoms now would have been caused
by emissions over the past 20 years, during which total vinyl chloride
production increased from about 10% of its current leveel.
Error Analysis
The following discussion attempts to estimate the errors involved
in this analysis other than the errors introduced by extrapolation from
high to low doses and from animals to man, which are discussed in
Appendix D. Quantitative estimates are provided below for the
following sources of error:
(1) Errors in the slopes of the dose-response lines due to the
limited sample sizes in the experiments.
(2) Errors in estimating the ambient concentrations around
plants.
(3) Errors resulting from not considering effects on the
population living further than 5 mi from plants.
(4) Errors resulting from not considering the decomposition
of vinyl chloride in the atmosphere.
The standard deviations of the estimates for the slopes of the dose-
response lines were calculated from maximum-likelihood large-sample
theory, as described in Appendix C. As mentioned above, the results
are 27% and 21% of the estimated slopes for liver angiosarcoma and all
cancer, respectively.
Ideally, an estimate of the errors in the diffusion modeling could be
derived from a comparison with actual monitoring data. Unfortunately,
such a comparison is not available. In its absence, we use the difference
between the two independent estimates as a measure of the degree of
uncertainty in the estimates. As was noted in Appendix A, the two sets
of results were generally within 25% of each other.
We can estimate the magnitude of the errors due to the 5-mile cut-
off and vinyl chloride decomposition as follows. Suppose that, say, 10
million people live between 5 and 10 mi from the plants where they
would be exposed to an average concentration of VC of about 2. 4 ppb,
based solely on the diffusion modeling. (The figure of 10 million people
is chosen to show the sensitivity of the calculation to the 5-mile cut-off.
The 5-10 mile annulus contains 3 times the area of the 5-mile circle,
but would probably contain less than 3 times the population, since in
many cases the bulk of the urban area would be within the 5 mi distance. )
This exposure would cause 3. 5 cases of cancer of all types per year and
corresponding numbers of other health effects, according to the linear
model. However, vinyl chloride is believed to decompose in the
-------
B-7
atmosphere with a half-life estimated at about 6 hours. Assuming an
overall average wind speed of 10 mi/hr, the average concentration
in the 5 to 10 mile annulus would be reduced by decomposition by
about 8%, so about 3. 2 cases of cancer would be occurring there.
At the same time, decomposition would reduce the average concentra-
tion within 5 miles as well, by about 3. 4%. The overall effect of these
compensating errors would be to increase the estimate by about 25%.
Assuming that the first two sources of error are independent, the
estimates can be combined using a Pythagorean formula. Thus, the
combined error estimate would be the square root of
(.21)2 + (.25)2
or about 33% of "the estimated 11 cases of cancer per year predicted
by the linear model. This is about 3. 6 cases per year. Since the
second two sources of error listed above are estimated to lead to an
underestimate by about 2. 8 cases per year, a total error estimate
might be plus about 6 or minus about 1 cases out of 11 per year.
{Note: under the log-probit model, the effect of the 5-mi cut-off
would be negligible. )
-------
Figure B-1. Comparison of linear and log-probit projections for
liver angio tar come (Maltoni experiment 8T 1)
+1 -
0.5
0.1
<•*
31
D
<£
-4,
1
1000
0.1
10
100
001
0.001
0 0001
Vinyl chlonda concentration (ppm)
-------
Appendix C. Mathematical Treatment of Dose-Response Data
The dose-response curves given in Appendix B were derived from
the animal experiments using an extension of the standard maximum -
likelihood (ML) method of statistical estimation. The results at
different doses are combined and the line chosen is the one for which
the likelihood of the observed results is the greatest. A discussion of
the standard ML method may be found in any elementary text on
mathematical statistics.
Equations for ML Estimators
Assume that M experiments (including the controls, if any) are
performed. In the i-th experiment, n, animals are exposed to a dose d2'
and Xi responses are observed. We assume that the response of each
animal is independent of the others, so each experiment is a set of
Bernouilli trials in which the probability of response p, is a function of
dt with unknown parameters. In particular, we consider two functional
forms here:
The method is readily adapted to other functional forms.
Eq. (1) will be called the "no zero response" case and is appropriate
when there is no measurable response at zero dose. Eq. (2) will be
called the "positive zero response" case and is appropriate when a
measurable response occurs at zero dose. (The mathematics allows
•x c 0 , of course, which would imply at threshold below which no
response would occur. Inferring the existence of a threshold is fraught
with difficulty. )
For the positive-zero-response case, the joint probability of the
results obtained is
(all sums and products are assumed to be over the range I - 1, . . . , M
unless otherwise indicated). The values of * and £ which maximize L
(unless they fall on the boundary C>+ 0< ^ 1 for some z ) satisfy
the equations
(1)
(2)
- TT (?;)(«(3)
iogL = 0
log L 0
(4)
(5)
-------
C-2
After simplification, Eqs. (4) and (5) reduce to
2,
Y, - U t- ) m t
6^ &dt )0-«-(?<()
o
(6)
-» (<*t$cL J yi.J.
& t-(3 eft )~(
0
(7)
There are a set of two simultaneous 2M-th order algebraic equations for
which no closed-form solution is possible. However, a numerical solution
can be obtained by, for example, Newton's method for two variables which
is described below.
For the no-zero-response case, Eq. (5) applies and is equivalent to
Eq. (7) with oc = 0. This simplifies to
This is, again, a 2M-th order algebraic equation which can be solved
numerically by the standard Newton's method, described below.
Asymptotic Variances
With the ML method, asymptotic variances of the ML estimators are
readily obtained. (See S. S. Wilks , Mathematical Statistics (Wiley, 1962),
pp. 379-81. ) These are given by
0
(8)
(9)
(10)
for the positive-zero-response case. Eqs. (9) and (10) reduce to
(11)
(12)
t- (?> ) (I - * - '3 di J~
-------
C-3
Again, for the no-zero-response case, the corresponding result is
obtained by setting oc. = 0 in Eq. (12), which gives
J.
>1,
V /L
(13)
Eqs. (12) and (13) were used in Appendix B to calculate the standard
deviations of the slopes of the dose-response lines.
Newton's Method for One Variable
We wish to solve the equation f(x) = 0, where f has a continuous first
derivative. Starting with an initial estimate xr, we approximate the
curve y = f(x) by its tangent at x = x0; the point where the tangent
intersects the x - axis is an improved estimate of the root.
The equation of the tangent is
y - f(x0 ) = f*(xc,) (x - xe) (14)
Setting y = o in Eq. (14) yields
fix.)
x = - fw*7) <15>
Eq. (15) can be iterated until sufficient accuracy is obtained.
Newton's Method for Two Variables
Here, we wish to solve the equations
f(x,y) = 0 (16)
g(x,y) = 0
and have an initial estimate xc , y0 . The plane tangent to f(x,y) - z = 0
at (x0,y0 , f(xc ,y0)) is
(fA , fy , - 1) • (x - xL, , y-yL , z-f) = 0 (17)
or
(x-xc, )fx + (y-y0 )fY - (z-f) = 0 (18)
where the subscripts indicate differentiation and all functions are
evaluated at (x0 , yc< ). Similarly, the plane tangent to g(x,y) - z = 0
at (x0 ,yc , g(xc , )) is
(x -xc)g + (y - y(, )gj - (z-f) = 0 (19)
-------
C-4
The planes given by Eqs. (18) and (19) intersect in a line, which in
turn intersects z = 0 in a point which is an improved estimate of the
root. This point can be obtained by solving Eqs (18) and (19) simultane-
ously with z = 0. This gives
£ - -- - 2- (20)
' * ^ 3 " S3*
U _ U ^ ^7 (21)
Eqs. (20) and (21) can be iterated until sufficient accuracy is obtained.
-------
APPENDIX D: USE OF HUMAN DATA TO ESTIMATE RISKS
In Appendix B the linear model describing the dose-response curve
for the incidence of liver angiosarcoma in rats was modified by two
factors in order to estimate the number of cases of human angiosarcoma
in human populations. These two (multiplicative) factors were 8.4, to
convert the 4 hrs/days, 5 days/week rat exposures to continuous
exposures, and 1/30 = 0.033, to convert the annual rates in rats to
in humans. In this appendix the available human data on liver angio-
sarcoma were examined to validate these factors, revised estimates of
the number of human cases per year of exposure to ambient vinyl chloride
concentrations around plants are made, and the number of people incur-
ring effects other than liver angiosarcoma is estimated. The emphasis
is placed on the quantitative comparison of data in rats and humans.
Interspecies Comparison of Life Expectancy and Angiosarcoma Latent Times
The most obvious difference between rats and people is their
relative size and attendant metabolic rate and life expectancy. Both
life expectancy and the time required for tumors to develop are independ-
ent manifestations of the basic differences in metabolic rates.
According to the 1970 U.S. census data, the life expectancy of one-
year old infants in the U.S. is 75.6 years for females and 68.3 years
for males. This is to be compared to about 100 weeks or 1.93 years for
rats. The ratio of life expectancies is therefore about 1.9/72 = 0.026.
The time required for tumor development in people can be estimated
from the case histories of 14 confirmed cases of liver angiosarcoma,
all of whom are highly-exposed workers in the vinyl chloride industry
(reference 2). The average time between first exposure to vinyl chloride
and diagnosis is 19.6 years, and the average duration of exposure is
17.2 years. The distribution of exposure durations preceding diagnosis
among these cases is approximately log normal, i.e. it is a normal
distribution when plotted against the logarithm of latent time. The
median time is 17 years, and the variance is 0.15 units of log (time)
per one standard deviation. By contrast, Mai torn" found that the
average time in rats between onset of exposure and appearance of tumors
was 80 weeks (after a 52-week exposure), regardless of dose between
10,000 ppm and 250 ppm. The one liver angiosarcoma which appeared in
the 50 ppm group occurred 135 weeks after onset of exposure, but this
one case is regarded as insufficient information to disprove the
conclusion that the latent time in rats is independent of dose. The
approximate ratio of times required for tumor development in humans and
rats is 1.54/1 9.6 = 0.078.
This estimate contains considerable error. One error is that the
-------
D 2
observed average exposure duration preceding diagnosis in the workers
ts an underestimate of the true biological latency. There are two
reasons for this statement:
a. The number of people in the industry has increased steadily
since 1940. Therefore the current work force contains an
over representation of people with short exposure times, and
the cases which have appeared are systematically biased toward
short exposure times.
b. Even if the number of exposed people had not increased in the
past, the disease could still develop in those people in the
future, which would result in a longer average exposure
duration before diagnosis.
A more fundamental difficulty in attempting a comparison of
exposure times for rats and humans is the uncertainty whether the
disease is caused by a gradual build-up of some type of tissue irrita-
tion which requires constant intake of vinyl chloride or whether it is
caused by an event occurring randomly in the exposure history and
progressing to angiosarcoma independent of subsequent exposures.
Intuitively, we think the latter mechanism may be happening in animals
because of the independence of latency time with respect to dose.
However, in people, there is evidence of latency time variation with
dosage (see reference 7). If there are indeed different mechanisms in
rats and people the ratio of exposure durations computed above has no
valid interpretation.
Latent time considerations would dictate that the annual
incidence of human cases should be 0.Q78 times the annual incidence of
rat tumors. The model in Appendix B assumed a factor 0.033, so that
exposure duration considerations would dictate that the rates calculated
from the model should be multiplied by 2.4.
One intuitive measure of the sensitivity of a species is to compare
for similar doses, the time it takes to get cancer, expressed as a
fraction of total lifespan. For rats this is 80/100 = 0.B of a lifetime
irrespective of dose and for humans it is 19.6/72 = 0.27 of a lifetime.
By this measure, humans are over three times more sensitive than rats.
Comparison of Angiosarcoma Incidence
The estimation of liver angiosarcoma incidence in workers is
possible from several epidemiology studies, but unfortunately the only
study in which it was possible to define the vinyl chloride exposure
(reference 4) had too small a sample size to detect liver angiosarcoma
cases.
In this plant, measurements of vinyl chloride sampled in the
breathing zone of the workers were begun in 1950, and continued until the
-------
D 3
present. It was found that in an old section of the plant, exposure
in the wet end operations averaged 120-385 ppm, over an 8 hour/day
5 day/week time period, with documented excursions up to 4000 ppm. The
measured excursions coincided with reports of workers feeling dizzy on
the job. In 1959, following reports of toxicological effects of vinyl
chloride, the operations were improved, a continuous monitor was installed
and subsequent exposures for the same job were reduced to 20-80 ppm with
excursions of 500 ppm. In a newer section of this plant the highest
exposure job was 135-825 ppm before 1959 and 30-240 ppm thereafter.
Since this is one of the largest companies and one of those most
concerned with worker exposure, it is likely that these are lower than
the industry average, so that one might assume from this that the cases
of angiosarcoma occurring among workers in the last 5-10 years were the
result of past chronic exposures of 200-500 ppm.
In Table 1 the incidence of angiosarcoma from several industrial
studies is summarized. The Tabershaw-Gaffey study (reference 5) of
records of 8384 men in the entire vinyl chloride industry classified
1817 of them in the high exposure, high duration category. Since the
study encompassed 33 plants, it was not possible to define objectively
what "high" exposure was, but instead each plant classified cases high
or low considering only its own experience. The term "high duration"
here means more than 60 months on the job. They found two angiosarcomas
in the 1817 people, but from detailed studies by other investigators an
additional four were found to be misclassified as other diseases, so
that the incidence of angiosarcoma among the high exposure, high duration
workers was found to be 6/1817 = 3.3 x 10-3. The average exposure
duration for this group was 15.9 years.
Nicholson et. al. (reference 6) studied 255 workers employed in
one plant for more than 5 years. Of these, 151 worked in direct PVC
production and 3 cases of liver angiosarcoma were found in this group
for an incidence of 20 X 10-3. They were exposed for 14, 17, and 23
years respectively. The exposure duration for the entire 255 workers
ranged from 5 to 28 years, and most of them were exposed for more than
23 years.
In the Heath and Falk study (reference 7) of angiosarcoma cases in
the Louisville plant, 7 cases were seen among 270 people working directly
in polymerization activities, for an incidence of 26 X 10-3. The mean
duration of exposure (employment) was 17 years in the seven cases. The
authors stated that these 270 workers represent about 20% of the plant
employees.
The study of J. K. Wagoner (reference 8) of 745 male workers in two
plants, 31 cancer deaths, 6 of them liver cancer were reported. This
population of workers had been engaged in the polymerization of vinyl
chloride for at least 15 years. The type of liver cancer was not
specified, so that this study can yield only an upper limit of the liver
-------
D 4
angiosarcoma incidence; it is less that 6/745 = 8 X 10-3. Since no break-
down of job categories was made in this study it is likely than many of
the 745 people did not have the highest exposure tasks.
In comparing the incidence in table 1, it is significant that
studies with the larger number of people have the lowest incidence and
the studies with the most complete definition of worker exposure have
the highest rate. It can be concluded that the incidence among just
the most highly exposed workers is about 0.02.
The incidence numbers in Table 1 represent the number of cases
diagnosed before 1974 divided by population exposed before 1974. These
numbers, however, do not really represent the true incidence of the
disease because all members of the population exposed before 1974 have
not had time to develop angiosarcoma because of its 17-year average
exposure duration. At least two alternative approaches are available to
adjust for latency time effects. The first approach described in this
paragraph assumes a fixed exposure duration and considers only the popula-
tion exposed earlier than this time. The second approach described in the
next section considers the entire exposure duration distribution. In the
first approach the incidence is expressed in terms of the population
exposed before 1974-17 = 1957, which is only a fraction of the popula-
tion exposed before 1974. To calculate this fraction, information
concerning the number of people entering the work force as a function
of time from reference 5, table 2 has been used. It is found that the
ratio of people whose exposure began before 1974 to the number whose
exposure began before 1957 is 3.8. Therefore the population in which
cases of angiosarcoma have appeared is 1/3.8 as much as the population
in table 1, and the incidence among workers is 3.8 times that given in
table 1, or 0.02 X 3.8 = 0.076. This means that sometime in their
lives, over 7.5% of all highly-exposed workers are expected to get liver
angiosarcoma.
The average exposure time for high duration workers reported in the
industry-wide Tabershaw-Gaffey report (reference 5) is 16 years. There-
fore, the incidence rate, defined as the number of cases per person-year
of exposure, is 0.076/16 = 0.00475 per person-year.
Alternative Approach to Calculating Incidence Rate
The difficulty with the previous approach is that a fixed exposure
duration was assumed whereas the actual distribution of exposure durations
is approximately normally distributed in -log (duration) for the 14
reported occupationally-exposed cases. For each year, t, of exposure,
n(t) person-years of exposure have occurred, and the number of cases,
c(t), that have occurred before 1974 due to the exposure at time t is
c(t) = n(t) p L(74-t)
where p is the probability per person-year of exposure that a person
-------
D 5
will get angiosarcoma sometime in his life after being exposed for one
year, and L(74-t) is the probability that the exposure occurred a long
enough time aqo to appear by 1974 if he was exposed in year t. The
function L (x) is the cumulative distribution of latency times and is
equal to the probability that the latent time is less than or equal to
X years. The total number of cases is the sum of c(t) over all times
from the beginning of exposure to times so recent that L (74-t) is
vanishingly small. For convenience we shall write n(t) as the product
of the total number of person-years of exposure* N, and a distribution
function, q(t) which is the fraction of the N person-years that occurred
in year t. With these quantities defined, the probability p can be
calculated from
For this calculation the function q(t) was derived from table 2 of the
Tabershaw study (reference 5), where the distribution of man-years of
employment is presented as a function of the year in which exposure
started. It is assumed that the distribution of man-years is the same
for highly-exposed employees as for all 7128 workers in his study. The
distribution of exposure durations for the 14 known occupational cases
of liver angiosarcoma was taken from reference 2 and used for L(74-1).
With these functions, the cumulative product in the denominator of
equation (1) is 0.378. This is the fraction of cases which have been
initiated by exposure through 1974 which have been diagnosed.
Therefore the overall effect of this more exact treatment, where the
exposure duration distribution is explicitly considered, is to multiply
the calculated probability by 1/0.378 = 2.65. This turns out to be not
much different than the earlier approach, where the factor was 3.8. We
consider the alternative approach to be conceptually more firmly based.
Therefore we obtain 0.02/(17 X 0.378) = 0.0031 as the probability per
year of exposure that a highly-exposed worker will get angiosarcoma some-
time in his life. The errors in this calculation are of two types:
(a) uncertainties in the numerical estimates of parameters; and (b) con-
ceptual errors in the assumptions of the model. The first type can be
quantified in the following way. The chronic dose experienced by highly-
exposed workers is estimated to be somewhere between 200 ppm and 500 ppm.
The value of 350 ppm was chosen as the mid-point of that range, so the
variation could be + 150 ppm, or +_43%. Although this error does not
strictly enter into the calculation of the 0.0031 value, it does contribute
to the error when this is compared with the animal data, as discussed on
page D-6. The incidence of angiosarcoma in highly-exposed workers was 0.02
(table 1) and by intuition the author would guess that it should be within
the range of 0.01 or 0.03, for a variation of 50%. The number of hours of
daily exposure in the workers is almost certainly not as much as 8, but may
well be over 6 hours, so one may say it is 7 + 1 or 7 +_ 14%. If we assume
that these errors are random, the error in the product of these factors is
P =
SL c(t)
in
N ^ g(t) x L(74-t)
-------
D 6
The conceptual errors in the assumptions of the model are not
directly quantifiable. One error is that we assumed no breaks in
exposure once employment began. People who left the job and returned
later have contributed their man-hours of exposure more recently than
we assumed, and would therefore not contribute as much to the occurrence
of cases observed today. This has caused us to over-estimate the
incidence. A second error is that, as discussed on page D-2, the
distribution of latent times is biased toward shorter exposure intervals
than is likely to be the actual case. The error acts in the same
direction as the first, and would cause a further over-estimate of the
incidence.
A third error is that the true biological latent time is not
directly observable, since the process leading to irreversable disease
may have been initiated some unknown time after the first year of
exposure. By using the duration of total exposure, we tend to over-
estimate the true biological latent time. This error operates in a
direction opposite to the preceding two errors. The net result of
these of these three errors is unknown in direction or size.
The overall conclusion is that the incidence rate is 0.0031 per
person-year of exposure with a quantifiable error due to uncertainties
in estimation of parameters of 67% and an additional unknown conceptual
error in the precise assumptions.
It is important to point out that the incidence rates calculated
in this Appendix are probabilities based on the duration of exposure to
vinyl chloride, rather than on the number of cases which happen to be
diagnosed in a population each year. In descriptive epidemiology the
latter concept is called the incidence rate. The epidemiological
incidence in the general United States population is discussed in
Appendix E.
Comparison With Animal Data
At this point we are in a position to compare the human annual
incidence rate (0.0031) with that predicted from the animal model in
Appendix B. For continuous exposure to 350 ppm, Appendix B would
predict an incidence rate of (0.071 X 10^) x 3.5 x 10^) = 0.0249 per
person-year of exposure. If this is adjusted to occupational
exposure conditions by dividing by (24 x l)f[l x 5) = 4.8, the incidence
rate for highly-exposed occupational groups is predicted to be 0.0052
per person-year of exposure. This is a factor of 1.68 higher than
the human rate. Therefore all of the estimates in Appendix B should
be multiplied by 1/1.68 = 0.60 to be consistent with the human data.
When the error estimates are explicitly taken into account, we find
for the animal model, that the range encompassed by plus or minus one
standard error is (0.0052 + 27%) = 0.0038 to 0.0066 and for the human
data it is (0.0031 + 67%) = 0.0010 to 0.0052. Therefore, within the
accuracy of our procedures, the two approaches for calculating the
incidence are in agreement.
-------
D 7
The rationale for comparing human incidence data with the rat
experiments is considered more firm than the alternative approach of
comparing latency times, as discussed on page D-l, because of the
conceptual difficulties already pointed out.
The other effects of vinyl chloride besides liver angiosarcoma are
evaluated in the remainder of this appendix.
Ratio of all cancers to liver angiosarcoma
The study of Nicholson, et al. (reference 6) showed that out of
255 workers 24 deaths were recorded, 9 of them were due to cancer, 3 of
which were liver angiosarcoma. Unfortunately this study cannot be used
to estimate the number of non-angiosarcoma cancers caused by vinyl
chloride because if the 3 angiosarcomas were not in the population, there
would be 6 cancers out of 21 deaths, which is not different than one
would expect in the general population. Reference 15, p. 2 indicates
that of all deaths in the general population, 1/5 of them are due to
cancer.
In the Tabershaw study of mortality in the VC-PVC industry
(reference 5) the cancer mortality of the high exposure group is only
slightly higher than the overall cancer mortality for the entire group.
Although this slight excess could be due to vinyl chloride exposure,
it is not enough to enable one to estimate how many cancers of all sites
are induced relative to liver angiosarcoma.
Studies by Ott, et al. (reference 4) and Wagoner (reference 8)
have shown statistically significant excesses in mortality from many
primary cancer sites, including brain, respiratory system, and lymphomas.
However, angiosarcomas were not observed, so that the ratios of
total/angiosarcoma incidence cannot be estimated from these studies.
In a proportional mortality study of 161 workers in the Louisville
plant, Monson, et. al. (reference 16) found a larger rate of cancers
than expected for U. S. white males at the following sites: digestive
tract (excluding liver), liver and biliary tract, lung and brain. They
found 41 deaths due to all cancers fcompared with 27.9 expected) for an
excess of 13 cases. In the same population there were 5 deaths due to
liver angiosarcoma (compared with a vanishingly small number expected).
Therefore the total number of cancer deaths is 13/5 =2.6 times the
number of liver angiosarcoma deaths. One cannot be absolutely sure that
all of the cancers in this population are due to exposure to vinyl
chloride, but in view of the other studies cited above, it can be said
that vinyl chloride is a strong risk factor in the development of other
cancers. In the rat experiments of Maltoni, a total of 76 animals had
Zymbal gland carcinomas and liver angiosarcoma combined, compared with
47 with liver angiosarcoma, for a ratio of total cancer/liver angiosarcoma
of 1.62. When all tumors were considered in the Appendix B, the ratio
was 2.0. Therefore, both the human and animal data show that vinyl
chloride induces about twice as many total number of cancers as liver
angiosarcomas alone.
-------
D 8
Frequency of Non-Cancer Effects Relative to Liver Angiosarcoma
In a literature review of 12,724 vinyl chloride workers in the
United States and Europe (reference 12) Marsteller, et al. found 118
cases of acroosteolysis (a gradual erosion of bone at the fingertips),
97 people with symptoms of Raynaud's syndrome (cold hands and feet),
40 people with skin legions, and 73 with liver disturbances. We might
assume that, of these 328 cases of reported symptoms, 20% of the people
had 2 symptoms each and the other 80% had one each, so that the total
number of people reporting symptoms in 328/1.2 = 273 people. This
incidence of overt symptoms (273/12,724 = 21 X 10-^) is tc be compared
with the incidence of angiosarcoma in a similar large worker population
such as is seen in reference 5, Table 1, which is 3.3 X 10~3. The ratio
is 7 people with symptoms for every liver angiosarcoma case.
In detailed clinical observations of 50 highly exposed people
currently working in plants, Marsteller (reference 12) finds that only
one had a history of liver disease and 8 had experienced Raynaud's
syndrome. Despite this lack of overt symptoms, he found that 38 of the
50 had pathologically high BSP retention times (a measure of serious
cellular liver damage), 31 had marked liver enlargement, which could be
palpated, and 42 had abnormally low blood platelet counts. Therefore for
every person with clinical symptoms of vinyl chloride exposure there is
expected to be 38/9 =4.2 people with liver malfunction serious enough
to cause abnormal BSP retention times. This brings the ratio of serious
liver malfunction to angiosarcoma to 4.2 X 7 = 30.
The frequencies of abnormal BSP measurements and low platelet
counts seen by Veltman (reference 13) in 70 vinyl chloride workers are
almost identical to those found by Marsteller, but Veltman found only
half the frequency of Reynaud's syndrome.
In a liver function screening program among 1183 workers at
Louisville, Creech and Makk (reference 14) used a large battery of
blood chemistry tests of liver function followed by more elaborate
diagnosis of people who were abnormal in the initial screen. In
this series of tests, 59 people were found to have liver function
abnormalities; 17 of them major abnormalities which required biopsies
and other elaborate tests. The plant physicians considered the 59
cases abnormal enough to justify moving the people to work areas where
there were no hepatic toxins. Of the 17 serious cases, 2 turned out
to be liver angiosarcoma. Therefore from this study, the ratio of
definitive liver toxicity to liver angiosarcoma cases is 59/2 = 30, a
ratio which agrees with the literature survey of Marsteller, et al.
(reference 12).
-------
TABLE 1 - LIVER ANGIOSARCOMA INCIDENCE AMONG HIGHLY EXPOSED WORKERS
Number of
Angiosarcoma Number of
Reference Cases People Surveyed Incidence (X IP"3)
5 6 1 ,817 3.3
6 3 151 20
7 7 270 26
8 <6 745 <8
-------
REFERENCES FOP APPENDIX D
1. "Statistical Abstract of the United States," 1!. S. Department of
Commerce, Bureau of Census (1974).
2. "Scientific and Technical Assessment Report on Vinyl Chloride and
Polyvinyl Chloride," EPA/ORD report, p. 143-145, June 1975.
3. "Annual Report of U. S. Vinyl Chloride Production," P. Tarasoff.
EPA Memo July 15, 1975.
4. "Vinyl Chloride Exposure in a Controlled Industrial Environment,"
M. G. Ott, et al, Arch. Environ. Health 30 333-339 (1975)
5. "Mortality Study of Workers in the Manufacture of Vinyl Chloride
and its Polymers," I. R. Tabershaw and VI. R. Gaffey, J. Occupational
Medicine 16_ 509-518 (1974).
6. "Mortality Experience of a Cohort of Vir.yl Chloride--Polyvinyl
Chloride Workers," W. J. Nicholson, E. C. Hammond, H. Seidman,
I. J. Selikoff, Ann. N. Y. Acad. Sciences, 246 225-230 (1975).
7. "Characteristics of Cases of Angiosarcoma of the Liver among Vinyl
Chloride Workers in the United States," C. W. Heath, Jr. and
H. Falk, Ann. N. Y. Acad. Sciences 246 231-236 (1975).
8. J. K. Wagoner, testimony at Senate Subcommittee on Environment,
Commerce Committee, 93 Congress, 2nd Session, Aug. ?1, 1974,
Serial No. 93-110 p. 59.
9. "Third National Cancer Survey: Incidence Data," National Cancer
Institute Monograph 41 (March 1975).
10. "Carcinogenicity Bioassays of Vinyl Chloride: Current Results,"
C. Maltoni and G. Lefimine, Ann. N. Y. Acad. Sciences 246
195-218 (1975).
11. "The Correlation of Clinical and Environmental Measurements for
Workers Exposed to Vinyl Chloride," C. G. Kramer and J. E.
Mutchler, Amer. Industrial Hygiene Assoc. Jour. 3^19-30 (1971 ).
12. "Unusual Splenomegalic Liver Disease as Evidenced by Peritoneoscopy
and Guided Liver Biopsy among Polyvinyl Chloride Production
Workers," K. J. Marsteller, W. K. Lelbach, R. Muller, P. ^ediok.
Ann. N. Y. Acad. Sci. 246 95-134 (1975).
-------
2
13. "Clinical Manifestations and Course of Vinyl Chloride Disease,"
G. Veltman, E. E. Lange, S. Julie. G. Stein, and U. Bachner, Ann.
N. V. Acad. Sci 246 6-17 (1975).
14. "Liver Disease Among Polyvinyl Chloride Production Workers,"
J. L. Creech, Jr., and L. '^akk, Ann. N. V. /*cad. Sciences
246 88-94 (1975)
15. "Cancer Rates and Risks," D. L. Levin, et al. , U. S. Department
of Health, Education and Welfare, 2nd Fdition, 1975.
-------
Appendix E
Is Residence Near Vinyl Chloride Plants a Risk Factor
In Frequency Of Deaths Due To Liver Angiosarcoma7
I. Introduction
Abundant evidence exists that vinyl chloride is one of the chief
causative factors for the high frequency of liver angiosarcoma among
highly-exposed workers in the vinyl chloride-polyvinyl chloride industry.
In proposing to regulate air emissions around these plants on the basis
of health hazards to people living near factories, it is important to
look for evidence that this disease occurs more frequently among
people living near plants than those living at random places in the
United States. Accordingly an investigation was made of the place of
residence, relative to the vinyl chloride and polyvinyl chloride
plants which are covered by the proposed regulation, of the people who
thus far are known to have contacted liver angiosarcoma.
II. Sources of Data
A. Plants and Locations
The list of plants and their addresses which was used in the
investigation is shown in Table I. It was compiled from the APHA
report (reference 1) and the EPA Air Programs Office document
(reference 2). This list includes some plants that are no longer in
operation and at least one plant which is reported (by APHA) to involve
no current work with vinyl chloride. Some of the plants make co-polymers
with vinyl chloride and other monomers. The list is believed to include
all plants in the contiguous states which make vinyl chloride monomer or
polyvinyl chloride resin. The fact that it contains extra plants does
not detract from this study; in fact it might be desirable to expand the
investigation to all plants known to fabricate plastics if it were not
for difficulty that such an expansion would excessively broaden the list
of chemicals which could be implicated as causal agents.
In several cases street addresses of the plants are not available.
In these cases it was assumed that the plant was somewhere within the
city limits of the postal address (for large cities) or was exactly at
the center of town (for small towns or suburbs with no definite boundaries).
B. Liver Angiosarcoma Cases
In September 1974, Dr. Henry Falk of the Cancer and Birth Defects
Section, Center for Disease Control (CDC) initiated an extensive case-
finding national survey of all liver angiosarcomas which have been
-------
E-2
diagnosed between 1964 and 1974. He obtained the information from three
sources'
(1) A systematic examination of death certificates filed at the
National Center for Health Statistics, Research Triangle Park, N. C.
Copies of death certificates were obtained for those deaths where the
diagnosis was similar to, or could easily be confused with, liver angio-
sarcoma. Death certificates always give date and place of death, age, sex
and race, date (and freauently place) of birth, residence and occupation
at time of death, occasionally the time lived at the residence, the
immediate and underlaying cause of death and a statement of whether an
autopsy was performed.
(2) Case records on file with the Armed Forces Institute of
Pathology. These included cases diagnosed at military hospitals.
Usually residence and occupational information were not provided, and
for the most part only age, sex, hospital location, autopsy number and
a minimum amount of clinical information were available from this source.
(3) Response to reouests for information originally sent out by
Dr. Falk to state public health departments and hospital pathologists
throughout the nation. Typically a letter from a state official would
list cases on record and Dr. Falk would send letters to the pathologists
identified by the state requesting further information and permission to
contact physicians and patients. This process invariably results in
complete clinical histories of the patients, (dates of hospital admission,
description of methods of diagnosis, detailed surgical and autopsy reports)
but rarely includes more than a bare minimum of personal information such
as residence and occupation.
Currently Dr. John Herbert at CDC is continuing the investigation by
interviewing next of kin and physicians in order to get residence,
occupational and medical histories, family members with cancer, possible
exposure to other chemicals through home hobbies, drinking and smoking
histories and any other personal information which could indicate causal
connection with the disease. As of early October, Dr. Herbert has
followed up about 20 cases in this way.
In order to confirm the initial diagnosis of liver angiosarcoma,
Dr. Louis B. Thomas, Director, Pathology Laboratory, National Cancer
Institute has agreed to examine the sections of tissue which CDC obtained
from the hospitals. This confirmation is regarded to be absolutely
essential, especially when attempting to draw conclusions from a small
number of cases. The information currently on hand is far from complete.
As of early October 1975, a large backlog of slides was awaiting confirma-
tion at the NCI and over 90% of the cases have not yet been followed up
by CDC.
-------
E-3
III. Procedures of this Study
With the cooperation of Dr. Herbert, the author, assisted by
Dr. Kenneth Cantor, went to CDC and extracted the available information
on liver angiosarcoma cases. The form shown on Attachment A was filled
out to the extent possible for each case in the files. To maintain
confidentiality of personal information, the names of the deceased were
not copied from CDC records. Instead the cases are identified in our
files by state and initial letters of the name.
A total of 286 cases were identified. Out of this total the cases
with information on place of residence at time of death were selected;
this group was further examined and cases with known or suspected
exposure to arsenicals or thorotrast and cases of known occupational
exposure to plastics fabrication and VC-PVC favrications were eliminated.
A subsample of the entire group of 286 cases was analyzed for age and
sex distribution and tested on a state-wide basis for clustering to a
greater degree than expected on the basis of population concentration.
Following the case selection procedure, the place of residence of
all selected cases was identified using maps and atlases of various
scales.
The initial survey of residences was done by locating cities and
towns in an encyclopaedia atlas (reference 3). This was found to be
adequate to locate residences to within 10 miles from plant locations.
In cases that were suspected to be within 10 miles of the plant, indexed
street maps were consulted at the Geography and Map Division, Library
of Congress, Alexandria, Virginia.
The plant locations had been identified on a series of U.S. Geo-
logical Survey topological maps (scale of 1: 250,000, or 10 statute
miles = 6.4 cm) by Teknekron, Inc. under contract with EPA's Office of
Planning and Evaluation. These plant locations were verified by detailed
street maps at the Library of Congress, for those plants where residence
of cases was less than 10 miles from the plants.
IV. Results of the Survey:
Of the 286 case records obtained from CDC, all cases from the
states alphabetically from A through M (a total of 166 cases where age
and sex is known) were analyzed for age distribution and sex ratio and
a test of gross clustering was performed. Such analysis for the entire
population of 286 cases will be performed at a later time. It was
found that of the 166 cases, 119 were males and 47 females, for a sex
ratio of 2.5 males/1 female. The age distribution for females is
approximately constant for all ages. It is more variable for males, but
there are not obvious bimodalities. When the male/female sex ratio is
examined at a function of age, it is found that for all cases less than
age 44, the ratio is about 1.0 (27 males/20 females), but for ages above
45 it is larger than a factor of three.
-------
E-4
An test was performed to test the hypothesis that the number of
cases in each state is proportional to the 1970 population of the state.
If the data should disprove the hypothesis, we would have evidence of
large-scale (state-wide) clustering of cases which cannot be explained
by population concentrations. The results were that (for the states A
through M) was no larger than chance variation would predict, >so that
there is no evidence of clustering on a state-wide scale.
Of the 286 case records obtained from CDC, 176 cases remained after
elimination of those with no residence information and possible thora-
trast, arsenical, and occupational VC exposures. The rejected cases
included 12 possible thoratrast exposures, two possible arsenical
exposures, one person exposed occupationally to plastics, and six cases
from Alaska and Hawaii.
After locating the residence cities and towns of the 176 cases
using an United States Atlas, there were only a handful of cases close
enough to the plants to justify searching the detailed maps at the
Library of Congress. Cases of distances larger than 20 miles from the
plant were only rarely recorded, and to date a complete tabulation of
number of people versus distance in the range of 5 to 20 miles of all
plants has not been made. However, for distances less than five miles,
there were only six cases, as shown in Table II.
Case number 1 of Louisville, Kentucky appears on this list through
a clerical error because by mistake his work address was traced and found
to be less than 5 miles from the plant, whereas the home address is the
subject of this study. This accident does cause one to speculate whether
a systematic investigation of employees working near vinyl chloride
plants would reveal some larger than expected rates of angiosarcoma. It
also emphasizes the importance of locating place of occupation in the
case of follow-up.
Cases 2 and 4 must be eliminated as candidates for vinyl chloride
induced liver angiosarcoma caused by living close to factories. Case 2
moved from Egypt to Jersey City just two years before his death. Cases
4 and 5 turned out to be other types of cancer, according to NCI path-
ologi sts.
It is still possible that Cases 1, 3, 5 and 6 can be attributed to
residence near vinyl chloride plants, but we will not know until NCI
confirms or denies the diagnosis, and until there is more complete
follow-up of the residence history of Case 6.
Case 6 is a special situation because he was an infant who died at
5 months of age from a liver cancer which is relatively common among
infants. This case must be eliminated from the formal study because it
is not a liver angiosarcoma. It could have been caused by vinyl chloride,
but we have no knowledge of any facts about the parents except residence
at the time of their son's death. Since transplacental carcinogenesis
has been observed in animal studies, it is possible that Case 5 is a
case of cancer caused by residence close to a vinyl chloride plant.
-------
E-5
V. Discussion of Results
Liver angiosarcoma is an extremely rare disease among the United
States population. According to data in the Third National Cancer Survey
(reference 4) the incidence rate averaged over the nation is 0.0128 per
100,000 population per year. At this rate, the expected yearly incidence
in the United States population (203 million people) is 26 per year, and
the expected incidence among the 4,580,000 people living within 5 miles
of the plants listed in Table I is 0.59 cases per year. Therefore for
the 10 year collection of cases in the CDC files we would expect 5.9
cases to occur within 5 miles of all plants if the presence of the vinyl
chloride plant contributed no risk factor pre-disposing people to the
disease.
The number of cases in our study, 286, is close to the expected 10
year total of 260, and the 3 possible angiosarcoma cases within five
miles of vinyl chloride plants would probably match well with the 5.9
cases expected if we could get residence information and trace addresses
for the cases in which we have no information. Since we could only trace
176 of the 286 cases, the approximate number of cases close to the plant
would total 3 X (286/176) = 4.9 if all were traced. For these reasons
our files probably contain most of the cases actually recorded in these
ten years.
The high male/female sex ratio and the striking manner in which it
increases above age 45 are indications that this is an occupationally-
exposed population, and there may be other factors in addition to vinyl
chloride.
This survey has produced no evidence that living around vinyl
chloride plants is a risk factor in the occurence of liver angiosarcoma.
This conclusion is far different than saying that living around plants
is not a risk factor for several reasons: 1) this type of survey of
a disease with a latent time from first exposure to diagnosis of 17
years reflects exposures that started at some time before 1957, when the
quantities of vinyl chloride produced were much smaller than the current
production levels. 2) This survey did not include the place of occupation
of the currently-suspected collection of liver angiosarcoma cases. There-
fore it underestimates the risk of being near a vinyl chloride plant.
3) This survey might not have detected all existing liver angiosarcoma
cases, although the number we have is consistent with the national
statistics. The circumstantial evidence for this is that there was not
substantial overlap between the three sources of case information. If
the data sources had been complete, the information collected by CDC from
the National Center for Health Statistics would contain all the cases
-------
E-6
reported by both the Armed Forces Institute of Pathology and the state
health departments. 4) In over 90% of the cases traced in this survey,
the only information available was the residence at the time of death,
and in some cases, the residence of the spouse or parent only. This
information is only a crude indication of where the individuals spent
most of their lives. 5) In contrast with our expectation when the survey
was started, there is a significant rate of changes in diagnosis after
the slides were confirmed by the National Cancer Institute. The 286
cases currently on file cannot be regarded as definitely-established
liver angiosarcomas.
In view of these limitations of the survey, positive statements
cannot be made about residence near plants being or not being a risk
factor. The only interpretation of these results that is justified is
that a search was made of the residence information which exists and this
search revealed no evidence that residence near plants is a risk factor.
In view of the long 17-year latent time it is expected that any survey
of current cases would underestimate the actual risk.
-------
Table I: List of VC-PVC Plants and their Locations
Company Type VC/PVC)
Location
1. Air Products and Chemicals, Inc. PVC
Pace, Florida
2. Air Products and Chemicals, Inc. PVC
Highway 95
Calvert City, Ky.
3. Allied Chemical Corp.
near River Mile 2355
Gulf State Road
Baton Rouge, La.
4. Allied Chemical Corp. VC
Geismar, La.
5. American Chemical Corp. VC,
PVC
Long Beach, Calif.*
6. Atlantic Tubing & Rubber PVC
Cranston, R.I.*
7. BAST Wyandotte
500 Central Ave.
S. Kearney, N. J.
8. Borden, Inc.
Geismar, La.
9. Borden, Inc.
Bainbridge, N.Y.*
0. Borden, Inc.
Compton, Calif.*
11. Borden, Inc.
Demopolis, Ala.*
12. Borden, Inc. PVC
111iopolis, 111.
13. Borden, Inc. PVC
Leominster, Mass.*
14. Borden, Inc.
near 59th & Interspace
Oklahoma City, Okla.
15. Continental Oil Co.
Lake Charles, La.
16. Continental Oil Co. VC
Old Spanish Trail
Westlake, La.
17. Continental Oil Co. PVC
Aberdeen, Miss.
18. Diamond Shamrock Corp. PVC
River Road
Delaware City, Dela.
19. Diamond Shamrock Corp. VC,
PVC
La Porte Highway
Deer Park, Texas
-------
Table I - P. 2
Company
Type (VC/PVC)
Location
20. Diamond Shamrock Corp.
Plaquemine, La.*
21. Dow Chemical Co.
VC
Freeport, Tex.
22. Dow Chemical Co.
VC
Oyster Creek, Tex.*
23. Dow Chemical Co.
VC
Plaquemine, La.
24. Dow Chemical Co.
PVC
Midland, Mich.
25. Ethyl Corp.
VC, PVC
Gulf State Road
Baton Rouge, La.
26. Ethyl Corp.
VC
Deer Park, Tex.
27. Firestone Tire & Rubber
Co.
PVC
Firestone Blvd.
Pottstown, Pa.
28. Firestone Tire & Rubber
Co.
PVC
Perryville, Md.
29. Foster Grant Co.
389 N. Main Street
Leominster, Mass.
30. General Tire Co.
PVC
Ashtabula, Ohio
31. B. F. Goodrich Co.
PVC
Walker & Moore Road
Avon Lake, Ohio
32. B. F. Goodrich Co.
PVC
2104 E. 23rd Street
Carson City, Calif.
33. B. F. Goodrich Co.
PVC
41st Street & Bells Lane
Louisville, Ky.
34. B. F. Goodrich Co.
188 Presidio Place
Williamsville, N. Y.
35. B. F. Goodrich Co.
VC
near River Mile 17
Tennessee River
Calvert City, Ky.
36. B. F. Goodrich Co.
VC
Henry, 111.
37. B. F. Goodrich Co.
PVC
Pedrickstown, N. J.
38. Goodyear Tire & Rubber
Co.
PVC
Plaquemine, La.
-------
Company Type (VC/PVC)
39. Goodyear Tire & Rubber Co.
40. W. R. Grace & Co.
41. W. R. Grace & Co.
42. Great American Chemical Corp. PVC
43. Keysor-Century Corp. PVC
44. Monochem, Inc. VC
45. Monsanto Co. PVC
46. Monsanto Co.
47. Morton Norwich Co.
48. National Starch & Chemical Corp. PVC
49. Occidental Petroleum Corp. PVC
50. Occidental Petrolium Corp. PVC
(Hooker Chemical)
51. Olin Corporation PVC
52. Pantasote Co. PVC
53. Pantasote Co. PVC
54. Pittsburgh Plate Glass VC
Industries, Inc.
55. Robintech, Inc. PVC
56. Shell Oil Co. VC
57. Shell Oil Co. VC
Table 1 - P. 3
Location
5408 Baker Avenue
Niagara Falls, N. Y.*
Owensboro, Ky.*
South Acton, Mass.*
FitchburgMass.*
2600 Springbrook Avenue
Saugus, Calif.
Geismer, La.*
730 Worchester Street
Indian Orchard, Mass.
Texas City, Tex.
Ringwood, 111.*
Meredosia, 111.*
River Road
Burlington, N. Y.
New South Road
Hicksville, N. Y.
238 S. Main St.
Assonet, Mass.*
26 Jefferson Street
Passiac, N. J.
Point Pleasant, W. Va.*
Columbia Southern Road
Lake Charles, La.
786 Hardy Road
Painsville, Ohio
State Highway 225
Deer Park, Tex.
Norco, La.*
-------
Table 1 - P. 4
Company
Type (VC/PVC)
Location
58. Schintech
PVC
Freeport, Tex.*
59. SCM Corp. (Huron Plastics)
Huron, Ohio
60. Stauffer Chemical Co.
2112 E. 223rd Street
Carson City, Calif.
61. Stauffer Chemical Co.
PVC
Delaware City, Del.*
62. Stauffer Chemical Co.
School House Road
Burlington, N. J.
63. Tenneco Chemicals, Inc.
PVC
Beverly Road
Burlington, N. J.
64. Tenneco Chemicals, Inc.
PVC
Flemington, N. J.*
65. Tenneco Chemicals, Inc.
VC
Painesville, Ohio
66. Tenneco Chemicals, Inc.
VC
4403 La Porte Road
Pasadena, Tex.
67. Union Carbide Co.
PVC
4337 McCorkle,
South Charleston, W. Va.
68. Union Carbide Co.
PVC
Highway 146 & Texas Avenue
Texas City, Texas
69. Union Carbide Co.
Taft, La.
70. Union Carbide Co.
Reported Closed
Niagara Falls, N. Y.*
71. Uniroyal Inc.
PVC
720 Fairport Nursery Road
Painesville, Ohio
72. Vulcan Materials Co.
Geismar, La.
73. Vulcan Materials Co.
VC
Deer Park, Texas*
^Location not included in APHA
estimate of total
population.
-------
ANGIOSARCOMA WORKSHEET
CDC #: Name: Sex:
Death Certificate: Place of Death: Date of Death:
Biopsy Performed: Yes / / No / / Date of Biopsy:
Autopsy Performed: Yes / / No / / Date of Autopsy:
Specimens Available: Yes / / No / /
Diagnosis:
(Immediate Cause)
(Underlying Disease)
Date of Birth: If not known: Age (at biopsy):
Age (at autopsy):
Age at Death:
Date of Diagnosis: Age at Diagnosis:
Residence at Time of Death:
Occupation at Time of Death:
Family: Married: Yes / / No / /
Children: Number: Ages at Death:
Pathology Reviews:
Specimens Received: Yes / / No / / Unavailable / /
Specimens Reviewed: Yes / / No / /
NIH Report:
AFIP Report:
CDS Report:
Followed up by CDC: Yes // No //
-------
CDC ?/: Other Patient I.D.:
Occupational History
Name of Employer Address From To Job Title Job Description VC As Other Chemicals
Residence History
City, State Address From To Distance from VC Plant Distance from Other Industry
embers of family exposed via occupation or residence? Yes / / No / / If yes, use additional sheet.
-------
Table II Case Summary
Case Identification
1. JLK, Louisville
Kentucky
2. IN, Jersey City,
New Jersey
3. HCM, Buffalo
4. CF, Niagara Falls
5. GB, Niagara Falls
6. RJ, Niagara Falls
Distance from Plant
Home: 6.7 mi. SSW
of B. F. Goodrich
Work: 4.3 mi. ESE
of B. F. Goodrich
Home: 2.0-2.4 mi. SE
of BAST Wyandotte
Home: Age Distance
0-20 7-9.2 mi
SW of B.F.
Goodrich
20-40 2.6-4.4 mi.
WSW of B.F.
Goodrich
40-58 8-9.5 mi.
SSW of B.F.
Goodrich
58-68 3.8-5.2 mi.
SW of B.F.
Goodrich
Home: 0.25 to 1.1 mi.
SSE of Goodyear
Home: 0.25 to 1.0 mi.
NNW of Goodyear
Home: 2.1-2-2.3 mi.
W of Goodyear
Time in Area
32-1/2 years
2 years
68 years
30 years
5 months
Age at
Death
33
41
68
63
5 mo.
41
NCI
Confi rmation
No
No
No
Yes, not
angiosarcoma
Yes, not
angiosarcoma
No
-------
Appendix E References
1. Landau, E., "Population Residing Near Plants Producing Polyvinyl
Chloride," American Public Health Association, Contract report,
EPA/OTS (August 1975).
2. EPA, OAQPS, Emission Standards and Engineering Division,
"Standard Support-Environmental Impact Statement, Vol. II
(draft)." April 1975.
3. Encyclopaedia Britannica, "Britannica Junior, Vol. 15" (1959).
4. National Cancer Institute Monograph 41, "Third National Cancer
Survey: Incidence Data," March 1975.
------- |