CARCINOGEN ASSESSMENT GROUP'S
PRELIMINARY REPORT ON
POPULATION RISK TO AMBIENT COKE OVEN EXPOSURES
This document is being released by EPA for external review
MAR 2 4 1973
ALBERT,M.D
CHAIRAN .,.
Elizabeth L. Anderson, Ph. D.
I. Nathan Dubin, M.D.
Charalingayya Hiremath, Ph. D.
Robert McGaughy, Ph. D.
Lakshmi Mi shra, Ph. D. ..
Ruth Pertel , Ph. D.
Wade T. Richardson, J. D.
Dharm Singh, OVM, Ph.. 0.
Todd W. Thorslund, Sc. D.
Adrienne Zahner, Ph. D.
-------
Introduction and Summary
As requested in a memo from Joseph Padgett dated April 5, 1977
the CAG.;.has. done an analysis- of the possible excess lung cancer risk
in populations living in the vicinity of coke ovens.. There is very
substantial epidemiological evidence that exposure to coke oven
; " ' *
emissions can induce- an excess risk of cancer in workers engaged in the
. production of coke or in the manufacturer of coal gas. This excess risk
is mainly due to a higher than expected death rate from lung cancer and
to a much lesser extent from a higher than expected death rate from'
cancers of-the bladder.,, prostate, pancreas and large intestine.
There are major difficulties in predicting- the cancer risk to
populations living in the vicinity of coke ovens. The chemical com-
position of emissions from-.coke ovens is exceedingly complex. These
effluents do contain known carcinogens, particularly those belonging to
the class of polycyclic organics. However, by analogy with cigarette
smoke, the overall carcinogenic effect of the mixture may be greater
than that ascribible to identified carcinogens. As the emissions move
away from the coke ovens, it is possible that the chemical composition
and the associated carcinogenic potency of the material may change
significantly. The best that can be done under- the circumstances is to
",''" - " - '"»
use an indicator such as benzene-soluble organics (BSD) as a guide to
the exposure of populations living near coke ovens. Unfortunately,
- there are no animal inhalation studies with coke oven emissions and
consequently no evidence on the extent to which dilution and aging
i
of emissions affects its carcinogenic potential or the most appropriate
indicator to use for carcinogenic effect.
-------
-2- .
Furthermore, there are no- epidemiological studies of the cancer
exposure of populations, living near coke,ovens or gas works as a
basis-.- for. estimating the: magnitude of the excess cancer risk,, although
such studies would necessarily provide only a crude measure of the
carcinogenic effects of coke oven emissions.
. s
In the analysis presented here,, we 'have used the data from a
series of studies by Lloyd, Redmond, Mazuiridar and co-workers- (1-4)
on; cancer mortality in relation to ESO exposure in coke oven workers
and other workers in the steel industry. Tnis epideir.iological dose-
response data for lung cancer is extrapolated on the basis of a linear
non-threshold/ dose-response relationship to .provide estimates of 'the
excess risk of lung cancer in populations living at various distances
from, coke ovens- The. exposure estimates and the size of the population
groups- are taken from an assessment by the Stanford Research Institute
(7). The validity of.the exposure assessment is outside the CAG's purview.
The risk assessment for lung cancer is summarized in Table 4. The
number of people exposed to coke oven emissions is on the order of
fifteen million. About fourteen million people have a lifetime excess
lung cancer risk of about 6 chances in 10,000. About one million have
a lifetime excess lung cancer risk of 1 chance in 1,000. The remaining
100,000 people have a lifetime excess lung cancer risk which ranges from
about 2 chances in a 1000 to 6 chances in 1000 depending on where they
live. Without any significant coke oven exposure, che lifetime chance
of dying, of lung cancer is 3.29%. For the 100,000 highest exposed people,
-------
-3-
there is a.:0.3% to.0.6£ excess chance of dying of lung cancer.
For., the. rase of the 15 million people the excess is about. 0.1%.
The total, number of excess, lung cancer deaths is about 150 cases
per year.
These estimates should, be regarded as crude and probably.
conservative; i.e., on the high sida.
-------
Estimated Risk Models Based -on Steel Workers'
Exposure Co Coke Oven Emissions
' 1-4
A. series- of studies- by. Lloyd, Redmond, Mazumdari and co-workers
has established a strong. relationship, between the. total exposure to coke
oven- emissions- and an increased risk of death, due to lung cancer.
'Land' in an OSHA hearing on coke oven standards used the most up-
f-
to-date summary of the data based on. the cited studies to estimate risk
to coke oven, workers. These data were supplied to EPA and. were further
summarized and adjusted' in- a manner shown and explained, in Table 1.
These-, data represent changing exposures to individuals over seg-
ments of their life spans with an observation of mortality over a dif-
ferent and, at least, partially overlapping additional segment of their
life spans. From this fragmented type of data, we wish to predict the
effect, of a constant exposure over the individual's entire life span to
the probability .of the individual'^ ultimate death being due to lung
cancer.
Weibull Model
To obtain a lifetime probability estimate, it is necessary to
relate the "instantaneous probability" of death due to lung cancer to
6
exposure and age. Following Armitage and Doll we assume that the instantane-
ous probability of death may be expressed as
h(t)
where t is the age of the individual, x is the level of exposure, and
Y, ct, & and a are unknown parameters to be estimated from the data shown
in Table 1.
The parameters were estimated using weighted nonlinear least
-------
_ 5 -
squares where the weights were taken equal to the person-year's of obser-
vation. This resulted in estimates
Y - 3..76934 '".-''' .
a - .27750 x 10"8
B-- .19248 x. 10~19'
*
ra = 1.21110
Oftea for theoretical reasons the parameter m is taken as an
integer, giving an approximate "m-hit" model. If we restrict m = 1, the
best fitting integer, we have
Y = 4.32324
a » ,2142 x 10"9
0 - .63192 x, lO'11
In Appendix I the lifetime risk of lung cancer and the expected
lifespan were derived, giving
n V»i (a + Sxm) _ a + PY o + PY
Q r») = = 1
2 '.(a + 6xm + PY) 1+_6jsL
a + pY
and
(a + Sx 4- p)
Y
respectively. The only unknown term is p , the term for non-lung-cancer
Y
deaths. We-can estimate p in the follox^ing manner. The median survival
time may be expressed as t where
-------
-6-
, 1/2'
pY)tY
or
pv
tf
x is, in this; case-, the national average ambient exposure. In the
SRI Report on coke oven exposure,, it. is- estimated that the 3SO Levels
arer
4.20 in cities containing coke- ovens
3.75 in cities not containing coke ovens
.95, in rural areas
Assuming- a locational distribution of the average American of
10%, 65% and 25%, respectively, in the three areas, the. national average
is estimated as x = 4.2(.l) -f- 3.75(.65) + .95(.25) => 3.095, so chat
65..36T
where 65.36 is the median life span of the U.S. nonwhite male based on
... 8 Y -9
1971. Vital Statistics. The term- p1 is estimated to be 9.60347 x 10 ,
-'-' 8 ':
when m = 1,. and 9.65501 x 10 when tn » 1.2111.
Substituting the estimated parameters into the risk and expected
lifetime equations, we obtain the results shown in Table 2, column 1.
However, this equation, is really strictly applicable only to a.
small subset of the U.S. population, namely the black northern raale who
was healthy enough to have a physically demanding job such as that of a
-------
- 7 -
steel worker. This.is true since these are essentially the characteris-
tics that the sample of workers in the epidemiological study possesses.
In making a prediction about the effect of coke oven emissions on the
U.S. population; in general, one can either use the. equation based on this
highly nonrandom sample or attempt to extrapolate the equation in. some
*
reasonable manner to the U.S. population.. ^
Method for Adjusting- Equations for One
Population to Apply to Another
For small exposures, x, the lifetime probability of a tumor may
be expressed as
a + p'
In, this: case the relative ratio of total to "non-BSO-caused" tumors can
be obtained by dividing -Q-C00) by a/ (a -I- p^) , giving
a
(
If we assume that'this ratio is constant between race-'sex-ragion, etc.,
then
a a
n
where S and a . are national rates. If we denote the national race as
n n
Q («) due to national average exposure of x , we have that
n n
-------
-8-
. '. ' 'a + B' xm' 1 '-t- 9x
^r\ - m -\s- m \
.a +&x+p L-h.9x4.p
a nan- . n n n
The Cera 0 («) can be calculated from.Vital Statistics by the segmented
n ..'.-' ' .
model discussed in Appendix I.. . Assuming Q 0=°) to- be known, we solve for
+
' ' - * .
the unknown
s-
a
n,
n
Thus the adjusted"lifetime probability of an individual randomly
selected from the U.S. population exposed to a level x of BSO may be
written as
- f , 1 + 9xm ' 1 + s + 1 -i- s
2 = 1 + 9xm + s = 8?
1 + s
Using 1971 Vital Statistics for total U.S. population, ICD Code
160-163, for respiratory system cancer rates, we calculated that Q C05) =
.0343. However, since lung cancer: ICD 162 makes up only a fraction,
-68,^17/72,898 = .9413, of the total respiratory cancer in 1971,. the
adjusted rate is estimated to be Q (°°) = .03"43 x .9413 = .0323. Thus
we .estimate s = (.0323* - 1)(1 + 3.095m9>, and the equations for the
national average are obtained by substitution of the appropriate terms
and are shown in Table 2, column 2.
Under the assumption that the fraction (E ) of total average life
span lived, given the same exposure, is the same for black males and the
total general population, we have that
-------
- 9 -
E = E. (a.+ 8xm + pY / a H-B-X + pY)1/Y
n. ; n
where x ..Is 3.095 and E is calculated Co be 70.96. The expected life
n . , . n,
span equations are derived by substituting the appropriate numerical
values into the equations and are shown in Table 3.
/ * '
»
Effects o£ Coke Oven Emissions on U.S. Population
. In the recent SRI Report, exposure of the U.S. population to coke
oven emissions was estimated. The number of people in an exposure gradi-
ent is given in columns 1. and 2 in Table. 4 and is. taken from the SRI
Report. The lifetime probability and the average length of life are
also shown in Table 4, based on the equations in Tables 2 and 3 where
the case m = 1, general U.S. population, was used. The other terms were
obtained by simple arithematic in an obviously straigbtforvard manner,
where length of a- life span is assumed to be 70.96 years. It is esti-
mated that about .2% of all lung cancer deaths per year are due to coke
oven emissions.
It would be possible to generate tables similar to Table 4
using the other three equations; however, the results would be only mar-
ginally different.
One potential criticism, of the use of the Weibull Model is that,
unlike other types of cancer,, the lung cancer rates have a tendency to
flatten out or even drop for older ages. However, while this does appear
to be the case in the low exposure groups, it does not in the highly
9
exposed group. Also as noted by Cook, et al. this flattening
appears to be explainable by a nonuniform historical exposure to envi-
ronmental .carcinogens and could be expected to. follow the power law in
the future if exposures were constant.
-------
- 10.
A: potential' method chat could be used to overcome this difficulty
is to assume that the total, lung cancer rates, h-Ct),. are made up of two
components: a, a rate, of an age cohort which is not Weibull; and 9 ,
which, is.. We write
where a is the nanexposed rate for the t interval and T is the average
d ' W
age of individuals, in the t interval. We are given 'the rates
.OgS)^-1
from 1971 Vital. Statistics tables so that we estimate
,. a. --r - YSCl.OgSoVf"1 ,
C. - - u. - - C
and the total rates given an environmental exposure of x are
h2t = Et + Y6Tj"1(xm - 3,095m) .
Using the constant segmented model discussed in Appendix I, the
increased lifetime probability where x = 10.9, m = 1, total U.S. popula-
tion, is.calculated to be 6.7 x 10 . This is_yery close to 6.04 x 10 ,
.
the value calculated for the full Weibull model. Thus it does not appear
that the fall'of cancer, rates for high age groups has a major impact on
overall lifetime probability of lung cancer.
-------
-XX-
TABLE 1
SUMMARIZED DATA, OF STEEL WORKER
EXPOSURE TO COKE OVEN EMISSIONS
Defined
Age at
Entry;
25-34
35-44
45-54
55-69
Group
Dose Range
Nb-C 'exposed
0-99
100-199
.200-299
300+
Npt exposed
0-149
150-299
300-499
450+
Not exposed
. 0-249
250-449
450-699
700+
Not exposed
0-249
250-449
450-749
. 750+
Lung Cancer
Rate/Year
x 105
13.4
31, 2r .
0. .
99.0
130.6
2.4.7
0
67 . 2
110.0
246.7
150.4.-
'S5-.5
234.5
258.9
601.5
70.0
203 . 7
167.8
558.7
2,222.2
Person-
years of
Observation
22,045
3,202
2,658
, 3,030
3,06.2
16,277
2,388
2,976
.2,727
2,027
11,306
1,527-
1,706
1,545
1,330
5,713
491
596
716
450
. Average Age
During
Observation
Periodb
36.4
,, 35.1
36.0
36.3
37.7
46.3
45.2
46.1
46.1
47.4
55.8
55.3
55.9
56.0
56.2
64.9
65.6
64.9
64.5
65.9
Life time-
Aver age
Exposure
to BSOC
4.2
20.5
67.2
109.1
160,5
4.2
28.4
82.3
130.1
201.5
4.2
47.9
107.4
169.4
263.3
4.2
43.7
89.8
160.5
681.7
Units, are the sum,, for all jobs, of the products of the mg/m of BSO
in the air associated with the job and the length of time in months worked
at the job.
The- term t is the average age at the start of the observation period
+ 1/2 total years observed -r total number of individuals alive at the
start of the observation period.
The term x is the total ng/m -months T total months lived by the end
of the observation period fraction of year spent on the job + 4.2 = the
background BSO levels in cities containing coke ovens. Units are
BSO.
-------
-12-
TABLE 2'
LIFETIME PROBABILITY OF LUNG CANCER GIVEN
AVERAGE EXPOSURES OF x Ugm/m3 OF ESO
Case
m = 1
m .= 1.2111-
Population
Black-Urban-Hale
.02164 + .0006438*
1 > .0006438x
.02794 + .0001938X1'2111
1 + .0001938X1'2111
General U.S. Population
.02966
1 +
.03147
1 + .
> .OOOSS24x
.0003824*
h .0002183K1'2111
0002183X1'2111
-------
TABLE 3
EXPECTED AVERAGE LIFE SPAN GIVEN AVERAGE
LIFETIME EXPOSURE TO x ugm/m3 OF BSO
Case
m 1
m i
«i 1 0 1 1 1
ra 1 . /Ill
Population
Black-Urban.-Male
i
109.907
93131
, (9.81589 + .0063192x) A
119.663
(9.93251 + .00192A8X1'2111)'26526
Total U.S, Population
120.408
OOl -11
(9,81589 t .0063192x) A
130. A92
(9.93251 -t- .0091248X1-2111)-26526
»»
co
I
-------
TABLE 4- ESTIMATED EFFECTS OF COKE OVEN EMISSIONS ON U.S. POPULATION UNDER
WEIBULL PROBABILITY KOOCL HMERE "HIT PARAMETER" nv=l MO ADJUSTMENTS
FCR TOTAL POPULATION RATES USED ' ,'
.X =
Exposure to
BSD irt/
-------
-15-'
References: .
1.. Lloyd J.W.,..-Lvindin F. E. Jr., Redmond C.K. al.: Long, term
mortality study of steel workers. IV. Mortality by work area..
J. Occup.;. Med. 12: .151-157. 1970 .
2. Lloyd J.W. r Long tern mortality study of steelv.-orkers. V.
Respiratory cancer in coke plant workers. J. Occup. Med. 13:.
53-68. 1971
3. Redmond C.K., Ciocco A:.,. Lloyd J.W..,. and Rush'II.W. : Long-term
mortality study of: steelworkers. VI. Mortality from malignant
neoplasms among coke overt workers. J. Occup. lied. 14: 621-629.
1972 .
4. Mazumdar S. , Redmond C.K.., Sollecito W. and Sussman N.: Aa
epidemiolog.ical study of exposure- to coal tar pitch, volatiles
among coke oven workers. J. Air Pollution Control Assn. 25:
382-389. 1975
5. Land C.E.: Presentation for OSHA hearings on coke oven standards,
May 4, 1976
6. Armitage, P. and Doll R: The age distribution cf cancer and
a multi-stage theory of carcinogensis. British J. Cancer, Vol 18:
: 1-12, 19\54 ..
7. .SRI draft report by Suta B.E.: Human population exposures to coke
oven atmospheric emissions, June 1977
-------
-16-
8. U.S. Dept. of Health, Education and "elfare: Vital Statistics of
. the United States, 1.971, Vol. IIMortality, Part A. Supt. of
Documents, U.S. Gdvt. Printing Office. Stock. No. Ol'7-022-00378-3
9. Goolc P.J.., Doll R. and Eellingham S.A.: A raathematicai model
for the age distribution, of cancer in aanr Int. J. Cancer, 4
' (1969),. pp 93-112
-------
- 17 -
COKE OVENS . .
. ;. .- . : .. .. APPENDIX: I
ESTIMATION OF 'LIFETIME RISK OF DEATH AND YEARS
OF LIFE LOST DUE TO A SPECIFIC CAUSE
General Approach *
The estimation of the lifetime probability of a disease in the
presence- of competing causes of death and the resulting life-shorting of
the disease is a problem that has received considerable attention. Chiang
(1968, pp. 242-68) has given a general solution to the problem using stan-.
dard methods in competing risk analysis. Gail (1975), using, these trethods,
derives explicit "figures of merit" for measuring the benefit of reduced
exposure to environmental carcinogens.
Following.Gail, we define the following functions:
S.(t) = P (survive to > t/c, is only cause of death}
S2(t) = P (survive to > t/c_ is only cause of death) ,
where c, and c~ make up all causes of death. The hazard function or age-
specific or instantaneous death rate is denoted as h (t), h (t), respec-
tively, for causes- of deathc, -and c^. Under the usual assumptions of
independence of c, and.c- the total probability of survival until time t
and the total hazard from both causes are: S(t) » S (t)S-(t) ,'
h(t) =h,(t) + h (t) , respectively and
SCO' = e
t
-/ h(u)du
o
-------
- 18-
It can be readily shown thac the lifetime probability of dying of
cause c_. is.
h2(t)S(t)dt
and the expected survival time or lifetime is
E' = / S,(t)h(t>tdt.
o '..
Thus, given the specific functions. Vu(t) , h.(t.), the. quantities of inter-
est Q2(°°) and E can be readily derived.
For two common assumptions concerning the form of h, (t), h_(t),
the terms Q-O") » E are obtained in the next two sections.
Constant Segmemfed Model
Often the hazard or age-specific death rate is assumed to be con-
stant over different time intervals but to change at-interval boundaries.
Formally, let us assume the entire life span is broken up into m mutually
exclusive segments or intervals and the hazard functions are defined as fel-
lows :
Time Interval
0 - t
: . ' 1
V^z
j-i j
t - »
m-l
^
a21
a'
?1
'V
h2(c)
a-
12
a22
a'
]2 .
\2
h(t)
a.
1
a2
a.
. J
a
m
-------
19 ~
'This is the situation where, age-specific rates are estimated from narrow
age intervals of typically 1 to- 10 years.
In, this case, the function S('t) , t. - < t < t. , is
' ' ' t"~C-T 1
-/ a.du -
S(t). = SCt.^) e ° = e.
so that
m
m
j-l
S(t - e
-(t.-t, ,)a.
°2i/aj
This.lifetime probability may be viewed intuitively as the sum of
the probability of m mutually independent events, namely of dying of the
specified, disease in the specified time interval. Each of these m proba-
bilities is the product of three probabilities:
1. Probability of death due to cause c. given that death occurred
in the j ^ time interval:
2.. Probability of death between time t. and t._. given survival up
to time t. , :
P2.;» 1' - e
; and
3. Probability of survival until time t.-:
-------
20
P3i-S(Vi) ' n C1'P2k)
3 J
By definition it follows that P_ = 1, P » 1,. and
The.expected life span under the. segmented model may be expressed
as ' . '''". . " .
-Ct-t. )o.
J J ^t-c )dt
-(t.-t. ,)a.l m
.= ?P /a
Weibull Model
Doll (1972) and others have pointed out that the Weibull distribu-
tion often fits time- and dose-dependent carcinogen response data. One
form of a- generalized Weibull model is to assume that
h2(t), = Y(u + Bx^'cT"1 , '
where t is the age of the individual, x is the level of exposure to a
carcinogen and y> ct, 8, m are unknown parameters to be estimated from the
data. If we make the additional simplifying assumption that the noncancer
cause of death c. is also Weibull in form with common parameter Y, then
-------
- 21 -
h1(t) -
and the quantities of, interest are
-/Y(a + 3x
(a + Sx^/Ca + Sxm + PT)
and'
E - /T(a * 3xm * PY)tV(a- * Sx + p )cY dt
o
-f. Bx"1 + Y1/Y
-------
-22-
References.:
1.. Chiang, C.L..:.. Introduction; to Stochastic Processes in
Biostatistics. New York; John Wiley,.1.968:
2.. Dolly R.:.. The age distribution of cancer: Implications for
model; of. carcinogenesis. J. Roy, Stat. Soc. A 134: 133-166, 1971
3. Gail, M.: Measuring the benefit of reduced exposure to
environmental carcinogens. J. Chron. Dis. 28: 135-147, 1975
------- |