Document qm08D31nVDjYDXrrGoZovn9p5

- . Risk Anuivsrs. Chi. 2, .Vo. 3, 1982 fl A Quantitative Estimate of Leukemia Mortality Associated with Occupational Esposure to Benzene' M a p C. White,' Peter F. Infante,2 and Kenneth C. Chu2.3 Recerced October ?8. 1981; recued Sepreniher 1. 198-7 ~-~~ ~ ~~ In 19Rn, the U S Supreme Court vacated a revlsed occupational standard for benzene. stating that the Occupational Sdfetv and Health Adrmmstrabon (OSHA) had twled to demonstrate that s i p f i c a n t health nsks e u s t e d under the current standard. Ths deci\ion has been interpreted by OSHA requiring the consideration of quantitative nsk assessments. ahenever possible. III the development of regulations for occupational carcinogens. In light of t h s deasion. the avarlable e p i d e m o l o g c evldence was used to generate a quantitaiirre nsk assessment for benzene. Uncertainties regarding the levels and lengths of benzene exposure for the s t u d e d cohorts were incorporated into the analvsis Based on the o n e - h t model. the assessment mdxates that a a o r h n g lifetime exposure to benzene at the current perrmsdde exposure level (10 ppm) poses a substantial excess nsk of death from leukema Ttus report, discusses the calculauon of the nsk estimates. the basis for relying o n c e r t a n asbumptiom, and the inherent h u t a t i o n s of using epidemoiogic studies to quantify cancer nsks LCEY WORDS: nsk aGeasrneni; benzene: leukema; occupational cancer; regulation 1. INTRODUCIlON The risk of developing leukemia from occupational exposure to benzene at the current. 8-hr tirneweighted average (TWA), permissible exposure level of 10 parts per d i o n (.pprn)(')has been \+gorously debated since the mid-1970s. when epidemiologic evidence conclusively demonstrated benzene's leukemogenic potential in humans. In 1978. the Occupational Safety and Health Administration (OSHA) promulgated a revised standard of I ppm T W A for occupational exposure to benzene.`" OSHA took the `The opinions. findmas. and blew expressed in this paper are those of the authors and should not be attnburcd 10 the Occupational Safety and Health Administration or the National Toxicoiogy Program. *Office of Carcinogen Identificauon and Classification. Occupational Safetv and Health Administration. 200 Consbtution Ave , N.W.R.oom N-3718. Wxdungton. D.C. 1U?IU. 'Curreni address: National Toxlcology Program. Bethesda. Maryland. position that "once the carcinogenicity of a substance has been established quahtatively, any exposure must be considered to be attended by risk when considering any given population." OSHA concluded that it was not possible to demonstrate a threshold level for benzene-induced carcinogenicity, or to establish a safe level for benzene exposure, and therefore decided that the permissible exposure level to benzene should be reduced to the lowest feasible level. The revised standard was challenged in the courts. and in July 1980. the U.S. Supreme Court upheld a lower court's decision to vacate the revised standard.`" A pluralitv of justices. i n a spiit 5 to 4 decision. stated that OSHA had failed to make the threshold finding that exposure to benzene at the current standard is unsafe. in the sense that significant risks are present and can he eliminated or lessened bv a change in practices. OSHA has interpreted the U.S. Supreme Court decision 3s requiring the consideration of significance of risk in the devel- 195 0272-4332 'tl2;090(t0195SOJ @.)/I r 1982 S n o e h for Risk Andvsir 1% W`hite. Infante. and Chu 4 opment of regulations for occupational carcinogens. includins nsk assessments \\hen thev can he Lippropnately performed. However. OSH.4 has stated that "when data are not available to perform a formal quantitative risk asscssment. qualitative evidence. expert testimony and C3ihc.r t.vidt.11c.e tiin\. h e appropriately utilized to based a determination of significance of risk."(4' Following the U.S. Supreme Court decision. the available epidemiologic evidence was evaiuated for use in a quantitative rish a i w s h m e n t . iincc n p p r n priate experimental data \rere i w t akaiiabic. Although benzene also has been shown to induce nonmalignant blood disorders, chromosomal aberrations, and perhaps lymphomas.'" attention was focused on the quantificativn ot the risk of death from leukemia. Risk assessments have been developed by the Environmental Protection Agency for environmental exposure to benzene@)and by the Consumer Product Safety Commission for consumer exposure to benzene.(7) These risk assessments were prepared with regard to regulatory considerations at the time and have not been published. Our purpose in publishing t h s risk assessment for occupational exposure to benzene is to stimulate interest and discussion on the methodology and assumptions which can be used when the underlying scientific evidence is less than ideal for the development of a quantitative risk assessment, in a scientific rather than a regulatory context. We invite critical evaluations of the relative merits of our approach to the epidemiologic evidence as compared to other approaches that could be or have been taken. 2. EPIDEMIOLOGIC EVIDENCE From epidemiologic studies, relative risk values and the corresponding exposure estimates may be used to develop a mathematical model for predicting carcinogenic risk. Most often in studies of occupationally exposed persons. the relative risk of a given cause of death can be approximated by the standardized mortahty ratio (SMR). However. because the number of workers who die from a particular cause of death in a population usually is small. the confidence interval surrounding the SMR may he quite large. In addition, adequate exposure measurements for the time period of interest may not be available. The error component of the mathematical model used to predict carcinogenic rick u-ill he ;I function 01 many fxtors. including the measurement errors asso- ciated Lbith the ShlR and e.xposure values. tiowever. measurement m o r in the exposure values IS an ejpe- ciaily i n i p o r t i n t problcrn. . I \ c`ipo\iirc 15 thc irit1c;li.n- dent variable of anv model. A large amount of' error in the independent variable could !cad to serious!y biased rewlts in the risk estimates. Thus. in order for an epidemiologic studv to contribute ubeful infcjrmativn t o the dc:\elnpmt.nt t > f a quantitauw risk assescrnent. i t mu\( meet t w o absolutely minimum requirements. First. the studv must provide an estimate of the excess risk in d population which \vas exposed to :he substancc in question. based on the expcnence of an appropria:e control population. Szcond. reliable industrial 11)- giene measurements n u t he available to permit a reasonable characterization of previous exposure con- ditions. The scientific literature is filled with case re- ports. case series. and epidemiolog~cstudies which qualitatively Link benzene exposure with leukemia in humans. This literature has been reviewed and sum- marized many How-ever. most of this sci- entific exidence has not Seen presen:ed in a manner that can be used to quantify the mazpitude of the risk of benzene-induced leukenua at a given exposure level. Only a handful of studies are available which attempt to measure the relative risk of lzukemia under specified benzene exposure conditions. Aksoy has published a series of reports which examined the incidence of leukemia among shoe- workers and other persons exposed to benzene in Aksoy reported that 31 shoeworkers had been diagnosed as having leukemia in one Istanbul hospital between 1967 and 1975, out of a population of appro?iimatelv 28.500 shoeworkers.""' Aksoy calculated a crude incidence rate of leukemia in this group of 13.Ci100.01X). kvhich uas more than double the leukemia incidence rate of 6i100.000 for the general population. This comparison does not take into consideration differences in ages between shoeworkers and the general population. the fact that persons diasnosed 3s having leukemia at other hospitals in Istanbul would not have been counted. or the absence of an accurate ieukeinia incidence rate for Turkey. Moreover. very little i s known ahout past hen- zene exposures among the 28,500 shoeworkers in Leukemia Mortality and Benzene 19 Istanbul. An early report by A k ~ o y ' i~n'dicated that the duration of exposure to benzene ranged from 7 months to 17 )ears for a sample of 217 apparenri!. healthv shoeworkers. and that the concentration of benzene in the working environment ranged from 15 to 30 ppm durin: nonworking hours tr! a peak of 2;O ppm when benzene adhesives were being used. Another report'"' indicated that benzene levels as high as 640 ppm had been measured in the shoeworking environment. The representativeness oi these few measurements to the entire shoeworking industw, on which the incidence rates are based, is impossible I O judge. Therefore, nsk values and exposure measurements based on Aksov's work were considered too speculative to be included in a quantitative risk assessment. In Italy, VigLani and hs co-workers also sought to assess the risk of leukemia among persons exposed to benzene. An early report('*)stated that 12 persons with leukemia attributed to benzene exposure had been diagnosed in the provinces of Milan and Pavia between 1960 and 1963: 11 had been diagnosed in 1962 or 1963. Assuming that about 5,000 people were exposed to benzene in these 2 provinces, Vigliani and Saita estimated that the incidence of acute leukemias in 1962 and 1963 was about 20 times hgher than expected. The investigators acknowledged that an analysis of the incidence of acute leukemia based on a control group would have been preferable. Even though there is considerable uncertainty in the relative risk of leukemia among these benzeneexposed workers, there is even greater uncertainty in the levels of previous benzene exposures. The only exposure information avdable is that benzene levels in some shoe factories in Pavia ranged from 25 to 600 ppm. with most between 200 and 500 ppm.(I3)I t is unknown how representative these figures may be of the exposure conditions of the 5,000 workers in Milan and Pavia who were estimated to be exposed to benzene. Thus, as Kith Aksoy's work, the information available from Vighani's reports concerning relative risks of leukemia and past benzene exposure levels were considered too imprecise to be used as other than a qualitative indication of benzene's leukemogenic potential. Only two epidemiologic studles were identified as meeting the minimum requirements for inclusion in a quantitative risk assessment for benzene. The first was a mortalitystudy of rubber workers in Oho. conducted by Infante and hs colleagues at the Sational Institute iar Occupational Safetv and liealth iYIOSH). Initial r e w l t of this ctudv :yere pbbil>ht.jin i97-.'ia.!7' 2nd A ~ i i o r &t t c i i k i F r e v n t a tion clf the SIOSH study was published recently hy Rinskv er al."" In [hi$ i t u d v . 748 white male workers were identificd u h o l i d been r x p ? ~ r (tio btxuene 11-1 tile p n 3 duction oi Hiotilm. a rubber film matenai. in two essentially icienticai production iaciIitie\. To he included in the studv, workers had to have worked for at leabt 1 dav between January 1. 1940. mii December 3 1, 194Y. in a department with benz.rne exp>wre. The number of *orkers who died between JmuarS, 1. 1950. and June 30, 1975. was compared io the expected based on U.S. white male mortality rates. using a modified Iife-table approach. With the Collow-up more t h m '907 cornplcte.-'5) the investigators found that 7 workers had died from leukemia. compared to an e.cpected fisure of 1.25. T h s excess in the number of workers who died from leukemia was statistically sigmficant (SMR = 560, P < 0.001). Six of these seven workers had been diagnosed with acute myelogenous or monocytic leukemia. In order to take into account the distribution of specific cell types of leukemia. an expected number of deaths was generated using mortality rates for acute ana monocytic leukemias from the National Cancer Institute."" The number of workers who died from acute or monocytic leukemia was found to be more than 8 times the expected figure of 0.70, yieldrng an SMR of 857. In the most recent report of this study.(16' Rmsky e t u l . reported that most of the 748 workers had been exposed to benzene for a relatively short period of time; 58070 of t h i s group had been employed for less than 1 year. When the data were analyzed by length of employment, a sigmficant excess in leukemia was observed among workers employed 5 or more years. but not among workers employed for less than 5 years. Among the latter group, 2 workers had died from leukemia compared to 1.02 expected. an excess which was not statistically significant. However, among workers employed for more than 5 years. 5 had died from leukemia cornpared to 0.23 expected, yielding an S h l R of 2100. The second epidemiologic study selected for this quantitative risk assessment \vas a mortality siudy by Ott et al. of Dow Chemical Company employees exposed.to benzene.('? A total of 541 white maies were 1% White, Infante. and Chu identified who had been exposed to benzene in 3 chemical production areas of a Michigan plmt, c.xcluciing ~ o r k e r ws ho ais0 had heen txposed ti, arsenicals, vinyl chloride, or asbestos. All workers were emplnved at the plant between 1940 and 1970, and no minimum length of emplovment was required for inclusion in this study. From 1940 to 1973, 91 workers had died. Three of these workers had died from leukemia; all were classified as myelocytic and two as acute myelmy-tic leukemia. Tl-us obsened numbcr was significantiv ( P < 0,048)greater rhan the cvpecled number o t c a s e ~ of leukemia of other than 1)rnphocytic or monocytic cell types, (essentially acute and chronic myelocytic leukemia), which was calculated to be 0.8 based on incidence rates from the Third National Cancer Survey. If we assume that the expected incidence of myelocytic leukemia was very similar to the expected mortality from myelocytic leukemia, then the relative risk of death from myelogenous leukemia among these workers can be approximated by an SMR of 375. 3. EXPOSURE CONDITIONS In recreating the exposure conditions of the workers included in both the NIOSH and Dow studies, an attempt was made to describe the range of average exposures. both by level and by duration. With regard to the NIOSH study, the initial conclusion of the investigators was that "employees' benzene exposure was generally below the recommended limit in effect at the time of each survev," (see Table I). During the 1977 OSHA rulemaking for benzene. some parties disputed this conclusion and suggested that benzene exposures may have been hgher than the recommended limits. The recent report of the NIOSH study(I6)presented a very detailed description of the processes and exposure levels at the studied facilities. The available information generally supports the NIOSH investigators' initial conclusion concerning exposure conditions. Although a few hgher exposures to benzene were recorded (up to 680 ppm). often these measurements were taken in areas where workers were present infrequently for brief periods of time or not present at all. Moreover. records at the plant indicated that the company was aware of benzene's toxicity and that respirators were worn during some operations. For t h s risk assessment, we assumed that exposures were at the recom- '"'"References prmidcd In infante er u/."*' and k n . n v e1 U I 'No rqcommended level was avrulahle beiore I34 I. `Recornmended lwei uas expressed 35 3 m;tx:mum a l l i m c ~ W t ~ concentrailon. Recommended level was expressed as a ceiling concentralion. mended standard for that time. We rccognizcd :hat some workers may have been exposed occasionally to higher levels. However, we believed 1hat the average exposure experience for the entire study population is adequately represented by the levels in Table I. and that the use of these levels actually may have overestimated average benzene exposures for many workers. The length of employment varied widely among the workers included in the NIOSH study. The minimum length of employment for inclusion i.n the study was 1 day, and the maximum length of time that anyone could have worked in one of the plants was 37 years. Well over half ( 5 8 % ) of the study population were employed for less than I year. In addition. the length of empIoyment for the 7 workers who died from leukemia ranged from 1 month to 20 years. with an average of 8.5 years. It was decided to base the risk assessment on the experience of workers who had been employed for 5 or more years, because most of the elevated leukemia risk was observed in this group. In addition, we did not know what proportion of the person-years for the total cohort had been contributed bv persons with only verv brief employment. The range of duration of employment was determined to be 5-30 years; the upper limit was based on the fact that workers who had been employed for more than 30 years contributed less than 0.01 to the number of espected leukemia deaths. The range of average benzene exposures was estimated as follows. The earliest date any employee was exposed to benzene was 1937, since this was the date the oldest facility began Pliofilm production. Workers could have begun emplovment as late as Leukemia Xlortalip and Benzene 199 1949 and still have been included in the study. Fol- low-up ended in 1975, so exposure to benzene dlter t h s date c d d not hake contributed t o the (kerved risks. Thus. workers employed for 5 vears could have been employed for any 5-year penod between 1937 and 1954. and \\corker$ empioved for 30 vearc could have been empioyed for any 30-yt.ar penod between 1937 and 1975. Because benzene-exposure measure- men:s were not available for the earliest years o f Plioiilni production. \+e arbitrariiy assumed that benzene exposures before 1941 were 50% higher. o r 150 ppm T W A . The average henzene exposure over ppm-yr. for our range of exposure. Therefore. the the period of 1937 to IY53. using 15U ppm for range of benzene exposures estimated for the Dow 1937- 1940 and the values from Table I. was 83 ppm. btudy was 42- 125 ppm-vr. I t should be noted that the Over the period 1937-1975, the average benzene estimated Cumulative doses for the three workers who exposure was 50 ppm. Thus, the exposure to benzene died from leukcniia were 545 pprn-mo (45ppm-yr), 19 whlch was associated with a 21-fold excess risk of ppm-mo (1.6 ppm-y). and 305 ppm-mo ( 2 5 ppm-y). leukemia was estimated to range from 83 ppm x 5 yr, Given that one-half of the study population and 2 of or 415 ppm-y, to 50 ppmx30 yr. or 1500 ppm-yr. the 3 leukemia cases ma); have had cumulative ben- Within the three production areas included in zene exposures of less than 42 ppm-yr, w-e believe the Dow study,(") the results of exposure monitoring that our range of 42-125 ppm-yr is unbkely to be an from 1944 to 1974 indicated that benzene exposures underestimate of the benzene exposure that was asso- were considerably lower than at the facilities studied ciated with a 3.75-fold excess risk of leukemia in t h s by NIOSH. Although peak levels of benzene as high study. as 937 ppm were measured. estimated time-weighted average exposures ranged from 0.1 to 35.5 ppm for different job categories. The length of employment 4. SELECTION OF A MODEL among workers in this study varied greatly; 23% were employed for less than 1 year and 17% were em- Several mathematical models have been devel- ployed for 20 or more years. Ott and h s colleagues oped to describe the relationship between the level of examined the job hstories of these workers and exposure to a carcinogen and the probability of de- calculated cumulative ppm-month career doses for veloping cancer associated with that level."*' We each worker. Their method was to multiply the mean selected the one-hit model for a quantitative risk T W A value for each job category by the number of assessment for benzene, primarilv because it is a months spent in that category. Information was not simple model. The current biological evidence con- prokided concerning the distribution of workers by cerning the chemical induction of leukemia was con- ppm-months. However. the observed and expected sidered insufficient to defend the use of a more number of deaths were analyzed bv cumulative dose complex model. In addition. the information avaii- (see Table 11). From this analysis, we determined able from epidemiologic studies regarding the relative that workers who had exposures of less than 500 risks of leukemia at different levels of benzene ex- ppm-mo, or 42 ppm-yr, contributed 5 7 8 of the num- posure was inadequate to statistically test the appro- ber of expected deaths from all causes. Workers who priateness of more complex models. had exposures of more than 1.OOO ppm-mo. or 83 The one-hit model is based on an assumption ppm-yr, contributed 298 of the number of expected that a single dose of a carcinogen can affect some deaths. From this information. we estimated that biological phenomenon in the organism whch subse- approximately one-half of the study group were ex- quently will lead to the development of cancer. This posed for less than 42 ppm-yr. and nearly one-third model is a nonthreshold model; Le., it assumes that - were exposed for more than 83 ppm-yr. The investigators did not state what the largest number of every level of dose is accompanied by some amount of excess cancer risk. ppm-months was for any worker in the study group. Expressed mathematically, the one-ht model We estimated an upper limit of 1.500 ppm-mo. or 125 states that the excess cancer risk ( P d ) is related to the 200 White, Infante, and Chu dose ( J )bv the equation If a "background" cancer risk (P,)exists in the absence of exposurc ta d. then P, i s not equal to the total probability of developing leuketma ( P,),but rather is related to P,and Po by the equation v,-P,,= Po)/(l- Po) (2) as explained bv Mantel and Bryan.''" Therefore, the excess risk ( P d )was redefined as P,,= [I -exp( - B x d )]( I - P,, ) (3) and the total leukemia risk as P, = p0 + [I -exp( - B x d ) ] (I - P ~ ) . (4) 5. CONVERTING INFORMATION FROM EPIDEMIOLOGICSTUDIESTO THE MODEL The relative risk of leukemia mortality from both the NIOSH('6) and Dow (I7) studies was ap- proximated by the SMR.The SMR represents the ratio of observed deaths due to leukemia in the benzene-exposed group. dibided by the number of expected leukerma deaths based on the death rates of a comparable age and sex-specific group SMR = number number of of deaths observed deaths expected x loo. For the risk assessment, the SMR is assumed to represent and thus, SMR = (P , / P o ) 100 P,= -S1(M0P0Ro). To determine the excess probability ( P d )of developing leukemia from a given level of benzene ( d ) over a defined period of time, using Eq. (3), two values needed to be calculated. The first \vas the working lifetime probability ( R ) of death due tu leukerma (all cell types combined) and myelogeneous leukemia, independent of ex- posure to benzene. The P,,values were calculated, using standard life-table methods. as the sum over 5-l;ear age intends (from 20 to 84)of the prnhabilirv of death due to leukemia during the specific interval times the prohahilit? of survival to the beginning of that interval. This life-table method assumes the independence of these events.("' Death rates for all causes and for all types of leukemia were taken from the 1975 mortality rates for U.S. white males."" Death rates for cell-specific leukemia were averase annual rates for U.S.wtute males from 1973 to 1977, and were abstracted from mortality rates collected through the National Cancer Institute SEER Program.'22' The calculated R , values were 0.00707 for ail types of leukemia and 0.00495 for myelogenous leukemia. (Appendicies providing detailed information concerning the calculation of these Po values are available upon request from the authors.) The secsnd value was B, which is associated with the rate at wtuch the excess probability of leukemia increased with each increment in dose. I t can be shown from Eqs. (4) and ( 5 ) that the solution for B is: B = -ln[(l-(SMR/lOO)P,)/(l-Po)]/d. (6) The SMR values and dose estimates which were used in these calculations are presented in Table 111. As explained earlier. upper and lower estimates of the dose associated with the SMR were generated for both the NIOSH(I6)and Dowti7)studies. Treating each study separately, the upper and lower dose estimates were used to calculate two E values from Eq. (6). These B values also are presented in Table Table 111. Values for SMR.Dose. and E" Used to Calculate Excess Leukemia R n k Studv SMR Benzene dose estimate B value corresponding to dose estimate Rinsky er 01.' 16' Ott era/."" 2 I00 4 I5 ppm-yr (lower) 1500 ppm-yr (upper) 375 42 ppm-vr (lower) I25 ppm-vr (upper) 0.000370 0.000102 0.OCXl?28 n m o i 10 "E is associated wth the rate at which the excess probabilitv of leukerma increases with each increment in dose. as calculated by Eq. ( 6 ) . Leukemia Mortality and Benzene 201 111. Separate risk estimates were calculated using each B value and Eq. ( 3 ) . The two resultant nsk estiniatss represent a range of nsk whch reflects the uncertainty in the exposure conditions for each study. 6. ESTIMATED EXCESS LEUKEhIIA RISK UNDER DIFFERENT EXPOSURE CONDITIONS Risk estimates were made for a working lifetime exposure to benzene, assumed to be 45 years. at both the current 10 ppm standard (450 ppm-yr) and at the vacated 1 pprn standard (45 ppm-yr). Recogwing that few employees would be exposed to benzene for as long as 45 years, estimates also were made for lengths of employment equal to 1. 5, 15, and 30 years at 10 ppm and at 1 ppm benzene. Table IV presents the estimates of additional lifetime leukemia risk (per 1,000 workers) due to occupational exposure to 10 ppm or 1 ppm benzene for different lengths of time, based on the NIOSH study. Table V presents the corresponding risk estimates for myelogeneous leukemia based on the Dow study. We believe that the NIOSH study provides stronger evidence of an association between benzene and leukemia than does the Dow study, as well as better information on benzene exposure levels. Therefore, we have greater confidence in the risk estimates based on the NIOSH study and our discussion will focus on these risk estimates. The risk estimates based on the Dow study are consistent with those from the NIOSH study. In fact, the risk estimates in Tables IV and V are much more similar than one m a y have expected, given the uncertainty associated with quantitative risk assessments. TaMc W.Estimates of Additional Lifetime Leukemia Risk per IO00 Workers" from Occupa- tional Exposure to Benzene* Exposure level 45 44-152 5-16 30 30- IO4 3- 11 I5 15-54 15-5 5 5-18 0.5-2 I 1-4 0.1-0.4 . *Estimates were rounded to the nearest whole number (per 1,OOO) for risks greater than 1 per I .OOO workers. b3asd on Data from Rinsky cr Table \ . Estimates of Additional LiIetime Leukemia" K J ~ Kper i(W Workers' from Oc<uDaii\?nai Lxoozurc to Uenzcne Fkponure levcl Years exposed 4s 30 15 5 I 10 pvm 4J- I !b 32-93 16-48 5- 16 1-3 1 vvm 5-15 3-10 2- I5 05-2 0 1-0 3 "Lrimdtes are b a e d on lrurrnua ceii t>pcrsother than l\mpoocvtic or rnoncr~~inc. 'Estimates were rounded to nearest whole number (per 1.ooO) for nsks greater than I per 1.O00 workers 'Based on data from Ott er d.'"' As can be seen in Table IV. 45 years of occupational exposure to benzene at 10 ppm was estimated to result in an excess lifetime leukemia risk of between 44 and 152 per 1.OOO exposed workers. At l ppm benzene for 45 years. the excess Lifetime leukemia risk was estimated to be between 5 and 16 per 1,OOO exposed workers. A certain magmtude reduction in either length or level of exposure produced a reduc- tion in risk of equal magnitude. The linear relations h p between dose and excess risk resulted from the fact that the one-hit model is essentially h e a r at low doses .'''' 7. DISCUSSION Because any risk assessment must apply study results obtained under one set of conditions to another and sometimes very different set of conditions. the estimated risk contains uncertainty. Although the best available scientific data were used in this risk assessment for benzene, certain assumptions had to be made for several unknown variables. The use of these assumptions might have resulted in a large amount of error in the risk estimate. So that the risk estimates are viewed in the proper perspective, the major assumptions on which these estimates were based are discussed below. First. it was assumed that the 748 white males in the NIOSH cohort and the 541 white males in the Dow cohort do not differ in their susceptibility to benzene-induced leukemogenesis from the more than 629.000 men and women, whte and nonwhte. who 202 White, Infante. and Chu are estimated to be occupationally exposed to benzene.(2' Even though this risk assessment was based on information derived from human populations, and therefore no species-to-species extrapolation was necessary. extrapolation from one population to another still may be problematic. Populations may vary from one another in their sensitivity and susceptibility to carcinogenic agents due to variations in genetic factors, age distributions, and exposures to potentially interacting environmental and social factors. However, no means are currently avadable to determine such variatian in susceptibility. With regard to age distributions, the SMR of a study population has been shown to be a useful approximation for the relative risk when the excess in mortality is consistent across all age group^.'^') It is not known whether the relative risk of benzeneinduced leukemia is different at different ages. However, if the relative risk of benzene-induced leukemia is age-dependent, then the SMRs obtained from the NIOSH or Dow cohorts could overestimate or underestimate the relative risk of benzene-induced leukemia among another group with a different age composition. Second, it was assumed that the risk ratios obtained from these studies represented working lifetime risks of benzene-induced leukemia. However, the risk ratios obtained from the NIOSH and Dow cohorts were based on exposure and follow-up periods which occurred over a limited fraction of the cohort member's Lifetimes. For instance. in the Rinsky et nl. report.('6) only 181 deaths were re- ported, meaning that 76% of the cohort was still alive or presumed to be alive at the end of the follow-up period. Similarly, only 91 deaths were reported in the Ott erul. study,('? leaving 83% of the cohort still alive at the time of follow-up. The assumption of a constant SMR could either underestimate or over- estimate the true risk, if the relative risk of leukemia among the cohort members actually increased or decreased during the period after the date of follow- up. . In addition, it also was assumed that the relative risk of leukemia would remain constant after ex- posure to benzene had ceased. Day and Brown'25) have demonstrated that, under the multistage theory, the effect of cessation of exposure will depend largely on whether the carcinogen affects an early or late stage of the cancer's development. If benzene were a late-stage carcinogen, then one would expect the ex- cess leukemia risk to eventually decline in these cohorts as the length of time since termination of exposure increased. However. PetotZ6)has argued that the multistage theory probably is not applicable to nonepitheliai cell cancers such as leukemia. Moreover, 3 of the 12 case reports described by Rinsky etaf.''6' were of men who had died from leukemia 9 or more years after they left the PLiofilm operations. To date. there is no evidence to suggest that benzene may be a late-stage carcinogen or .that the excess leukemia risk will decline after exposure has ceased. Third, many of the industrial processes where benzene exposure occurs, such as petroleum refining, rubber manufacturing, and chemical processing, contain other known or suspected carcinogens. The presence of other, carcinogens in the workplace might enhance (synergistically) the carcinogenic effect of benzene. However, it was assumed that no such synergism had occurred in the workplaces studied or would occur in other occupational settings. Fourth, one of the most important assumptions made in this risk assessment was that the doseresponse relationship between benzene exposure and leukemia development follows the one-hit model. Although this model is generally considered to be conservative at low exposure levels, it may actually underestimate the risk at lower levels if the risk of leukemia no longer increased with increasing dose after a certain level of exposure, i.e.. the dose-response curve plateaus, and if the studied workers had been exposed to benzene at a level above the point at which the curve flattens out. Fifth, it was assumed that for benzene exposure levels which are generally not associated with acute toxicity (TWA less than 100 ppm), a cumulative dose-effect relationship exists. That is, the level of risk associated with a certain dose was assumed to be constant. regardless of whether exposure occurred over a short period of time or at lower levels over a longer period of time. At present, there is no available method to test the validity of this frequently made assumption. The cumulative dose model also assumes that the total amount of benzene to which a worker was exposed contributed to his risk of death from benzene-induced leukemia. Theoretically, a worker could be exposed to benzene after the point at which the irreversible but subclinical development of benzeneinduced leukemia had begun. Thus, any exposure to c .q . Leukemia Mortality and Benzene 203 benzene after leukerma had been induced would not suggest that workers exposed to benzene at the cur- add to the risk of mortahty from leukema and could rent OSHA perrmssible exposure level oi 10 ppm are be mewed as a "wasted" dose in terms of generating at a substantially elevated nsk of death from leukemia. a dose-response curve. Therefore, a dose-response relationship which is based on total dose could over- estimate the amount of dose that is responsible for a given level of excess risk. REFERENCES Sixth. although these risk estimates take into consideration some of the uncertainty in the benzene levels from the NIOSH and Dow studies. these estimates do not reflect the uncertamty associated with I . OSHA Safety and Health Standards for General Industry, 29 CFR 1910.1000.Table 2-2. 2. Occupatlonal Safety and Health Administration. Occupational exposure to benzene: permanent standard. Federul Regtsrer 43. 5918-5970 (February IO. 1978). the risk ratios from these studies. For instance. the 3. Inclurtrial C h o n Department L'. American Petroleum Instirwe. upper 95O?o confidence interval for the SMR of 2100 from the Rinsky er al. report1"'l is 5073. The use of 448 U.S. 607 (1980). 4. Occupational Safetv and Health Administration. Identifica- tion, classification and regulatlon of potentld occupational this value would have resulted in much higher risk carcinogens: conforming delctlons. Federal Register 46, estimates. Thus, even though these risk estimates are expressed as ranges, the upper range of the risk 4889-4893 ( J a n ~ q19. 1981). 5. B. D. Goldstein, Hematoxicitv in bumans. in "Benzene tonic- ity: A Critical evaluation." Journal of Turicdogv and Enciron- estimates actually would have been greater if the mental Health. S. L a s h and E. D. Goldstcin, (eds), Suppl. 2. uncertainty of the SMRs had been included in the analysis. 69-105 (1977). 6. R. Albert, E L. Anderson. C. Hiremath. R. McGaughv. S. Miller, R. Pertel. W. Richardson. D. Sin&. T. W. Thorslund. and A. J. Zahner. "Carcinogen Assessment Group's Final Rcport on Population h s k to Ambient Benzene Exposures." U.S. Environmental Protection Agency, Washmgton, D.C. 8. CONCLUSION (1979). 7. P. D. White. M. S. Cohn. and W. K. Porter, Benzene Risk The risk estimates generated for total leukemia Assessment (U.S. Consumer Product safety Commission. Washtngton. D.C.. 1981). from the NIOSH studyciha'nd for myelogeneous leu- 8. National Research Council. Assembly of Life Sciences. Com- kemia from the Dow study`"' were almost identical. We estimated that a working lifetime exposure of 45 mittee on Toxicology, Healrh Ejjecrs of Ben:ene: A Review (National Academv of Sciences.Washmgton. D.C..June 1976). 9. M. Aksov. K. Dincol, T. Akgun. S.Erdem. and G. Dincol. years to benzene at 10 ppm (the current OSHA Haematologcal effects of chromc benzene poisorung in 217 standard) would result in a leukemia risk r a n h g from approximately 44 to 152 excess leukemia deaths workers. Bntish Journal of Indurtrial Medicine 28, 296-302 (1971). IO. M. Aksoy, S. Erdem. and G. Dincol. Types of leukemia in per 1,OOO exposed workers. A working Lifetime ex- chronic benzene poisoning. A studv in t h y - f o u r patients, posure to benzene at 1 ppm would result in a leukemia risk ranging from 5 to 16 excess leukemia deaths per 1,ooO exposed workers. At different lengths of ex- Acta Huemarolopira 55. 65-72 ( 1 976). 1 I. M. Aksoy. Different types of malignancies due to occupational -exposure to benzene: A reviewof recent observations in Turkey, Environmental Research 23. I8I 190 (1980). posure, the risk associated with 10 ppm was con- sistently 10 times greater than the risk at 1 ppm. This 12. E. C. Vigllani and G. Saita. Benzene and leukemia, New England Journal of Medicine 271. 872-876 ( 1964). 13. E C . Vigllani. Leukema associated with benzene exposure. order of magnitude difference in risk was expected Annals New York Academy of Sciences 271. 143-151 (1976). because the risk estimates were based on the one-hit model, and the one-hit model has beem shown to be 14. P. F. Infante. R. A. Rinskv, J. K. Wagoner. and R. J. Young. Leukemia in benzene workers. Loncet ii. 76-78 (1977). 15. P. F. Infante. R. A. Rinsky. J. K. Wagoner. and R. J. Young. essentially linear at low doses. Letter to the editor. Lancer ii. 868-869 (1973. This report demonstrates how some types of epidemiologic evidence can be used to estimate the level 16. R. A. knsky, R. J. Young, and A. E. Smith. Leukemia in benzene workers. Amencan lownu1 of Industrial Medicine 2. 2 17-245 ( 198I). of Carcinogenic risk associated with a certain level of 17. M. G. Ott. J. C . Townsend. W. A. Fishbeck. and R. .4. exposure. However. it is far easier to quantify a level of carcinogenic risk than it is to quantify the level of Langner, Mortality among indniduals occupabonallv exposed to benzene, ,4rch1cesoj Owironmental IIwlrh 33. 3- 10 (1978). 18. A. Whittemore. and J. B. Keller. Quantitative theories of - confidence that should be placed in the risk estimate. When viewed in the context of scientific evidence which qualitatively indicates that benzene is a rela- carcinogenesis. SIAM Rmrew 20. 1-30 (1978). 19. N. Mantel and W. R. Bryan. "Safetv" testing of carcinogenic agents. Journal of the National Concer Irrrtiture 27. 455-470 (1962). tively potent human carcinogen, these risk estimates 20. M. Gail, Measuring the benefit of reduced exposure to en- 204 vironmentd carcinogens. Journal u/ Chronic Diseares. 28. 135-147 (1975). 21. U.S. Sationai Center for Health Statistics. b ' d Siurtsrrtlr o j . ihe L'ntrrd Siutes. I975, Volume I I -Morrulirv ( US. Depart- ment oi Health. Education, and Welfare. Hyattsvlile, Mary- land. 1979). 22. J . L Young. C L. Percv. A. J. Asire. J W'. Berg, M. M. Cusao. L. A. Gltxcblsr. J W. Horm. U' I. Lourie. E. 5. Pollack, and E. M. Shambaugh. "Cancer Incidence and Morcahtv in the United States. 1973-1977," National Cancer Institute Monograph *57. NIH Publication No. 81-2330, Bethesda .MarylandiJune 1981). 23. D.G. Hoei. D. W. Gaylor. R. L. Kirschstein. U. Saffiotti. and White, Infante, and Chu . \ bl. A Schneiderman. Estimation of nxk of irreversiblc. delawd t o w i t \ . Jnurnul(?i To.xi,.slcolou~ncl Eirt;rronnrmrulllruiih 1. 133- 151 (1975). 24. M.J. Svmons and J. D. Taulbee. Practical considerations for approximatmg relative nsk bv the standardized mortahv ratio. Journul u j Occupurronul Medicine 23.4 13-4 16 ( I Y8I ). 25. N. E. Dav and C C . Brown. 5lultistaee models and pnmiq prebrnbon oi cancer, Journal v j the i\~uriunul c'unrer lrurrrure 61.Y77-YRY (19x0). 26. R. Peto. "Epidemiologv. multistaae models. and short-term mutagenicitv tcsts." in Oriqrnr of IIuniun Cdncer. 1'01.4. H I f . Hmt. J. D. Watson. and J. A. Winsten. eds. (Cold Spnng Harbor Laboratory, New York. 1977. pp. 1403- 1 4 3 ) .