Document npN8NdJD3372we9n70Mz2zDaX

Letter to the &&or Tackling a Very Difficult Problem H.G.S. Van Raalte, P.Grasso, and D.Imine We were interested to read the article by White er al. published in Risk Anabsis. The authors made a commendable attempt at tackling a very difficult probiem, but that an a number of poinu we would like to d c , particularly in relation to the estimation Of leveb ShCC M y arOn blUOduced hthCSC estimations an bound to have a major impact on the sscssmmt of risk at low lcrcls of benzcnc exposure. Ar the authors themsclnsadmit, exposun lcvels in the plants Jekctcd for study by Infante er al.") were not w d documented prior to 1951. and at the location where five out of Seven leukcma deaths occurred, there were no exposure data prior to 1957. Thus, the actual aposures of the lcuktmia cases during the a t t i o I ~ & Im~eanin@ period (193919S1) an largely unlaown.There are however some hdications that amcmtrations inside the spreader and dryer units have bcea occasionaily measured and indicated "a b@y dangerous levcl of benzene, rang ing from 200 to 350 ppm" to which anployas might have betn exposed whik making adjustments to the machinay. Although employctswtrc fcqucd to wear protective equipment when performing this task,this requirement was frequently At the other location where two deaths occurred, thae were no expure data until 1946 when an extensive zxhaust system was installed. From 1946 to 1950 only 15 mcasuremcnts were taken on four different days. Of these, 11 (73%) were above 35 ppm, the standard in effect at the time. There is also additional evidence of exposure in exctss of 500 ppm in certain instances. Thus, in our vim, the authors' "arbitrary" assumption that bcnzene exposures before 1941 were on "average"50% higher than those during the period 1941-1946 is based on very poor data indeed, and we consider it to be most unreliable to form the basis of a risk estimate. It is highly questionable indeed whether the "average"concentration of benzene, even if bascd on reliable data, can be considered as the only, or even the main, relevant ex~osunfactor associated with the and Wtto formed the buis of the risk andysu were employed in occupations whicb made arposures to "peak" lmlsvay likdy indecdal The imporunce of peak exposurt levels is ah0 apparent in the study conducted by Ott er al. In t b ~ ~ study, the authors collccdc that pak wrpc#urt Ievels of 937 ppm werc wdsurcd. We are not howeva told how frequent was tbc atposurt to these high lmls nor how many workmen were arposed to tban. S i a wc bold that CXPOJU~ to ~0nccnVati01is~CN- dal dC-8 t h e m Of h k d &wt f d that important and relevant informtion is missed when the mtbanatialmodellingis bued on a aunuhtive dose model. The authors mention, but do not include in their analysis. the papas by Askoy('' and Vighani.(s' De- spite the^ dtficiacia t h e anrt the fint publica- tions to indicate that benzene was leukemogenic to man. They also indicate that for the leukemia to occur, GX~OSWCIO benzene vapou!rs has to hc sub- stantial. This is in keeping with what we believe the study by R u d y r n and by Ott') d y demonstrate. With regard to the assumption that a cumulauve dose relationship exists for &ne, we would like to point out that although thm is, strictly speaking. no acceptable means for testing the assumption, it would not seem to accord with what is known about the toXicology of benzene. In fact an analysis of epide- miological data concerning the incidcna of leukcma and marrow hypoplasia suggests that both thcse con- ditions occur in persons exposed to levels of benzene 1 2 REFERENCES I. P. F.Infante, R. A. Rinsky. J. K. Wagoner. and R. J. Young, " L c u k e n ~ rLO B e m ~ Wa ~O~~CCSt.o"ncef X 76-78 (19771. c Letttr to the Editor Benzene D.Inine' The authors have Carried out a useful exercise in ah attcmpt to demonstrate how existing epidexnb logical studies can be utilized in estimating the risk of lcukcmia after long term exposure to low levels of benzme concentration. Ibe authors mention in detail many of the asuunptions they have had to make and give plausible arguments to counter many of the objections that one may raise. However,before one can accept the cxtrapolations that they produce, some further points need to be addressed. Fitstly, no matter how the data of the Rinsky('' and 0ttc2'studies arc modelled mathematically using a cumulative dosc model, large CXC~SSCS of death due to leukemia arc suggested at low levels of exposure. Sicc this is the case. the natural conclusion should k thrt if thcse two studies arc representative of baucaccxposedworkers worldwide then potentially large ucessts of lcukcmia deaths will result with prolonged follow-up. Before we accept this reasoning we must determine whether the two studies do in fact reflect typical working exposures of industrialized populations to benzene. The studies are special in the sense that complicating interactions with other chemicals do not occur. but an attempt at quantifying exposure, by necessity, concentrates on the available data and produces 'average' valucs. 'Ihe exposures during the early period in the Rinsky study are not documented but one surmises that the situation was much worse than the period after installation of extensive exhaust equipment, with a potential for temporary exposure to high concentrations. This point is of particular relevance in the matbematid modelling of the dose response rela- 'The British Petroleum Company, Chcrtscy Road. Sunbury-onTluma, Middlesex TY I6 7LN. England. tionship since a heavy reliance on a cumulative dose, ignoring the possibility that high exposures, albeit for brief periods, may be IIcceSsary for the induction of the leukemia, d o produce completely erroneous results in an extrapolation to lower cumulative doses. Tbae is nothing to be gained hac by simply objecting to this modelling since the biological knowledge is not sufficient to proposc any dternativc. What is csscntid is some further information concerning the shapeof the dose rcsponse curw in the 10- dosc region. One can either produce negative studies as in the case of the OSHA hearings, as counter exam- pits (albeit with thtif inhmnt defrci~ci~-Whitt er al.)"' demonstratrng that the projected risk values from thesc two studies do not hold, or one can attempt to see if the d t s do apply to actual populations. Adopting this latter approach there arc two gtoups of workas that can be cited who provide a good framework on which to judge the applicability of the risk ath8tts. The first corresponds to the "dry side'' workers referred to by Hannis''' who apparently did not satisfy the cohort definition applied by Infante in his original study. It has been stated that this group of 404 worken were exposcd to up to 20 ppm lmls of benzene, and yet by the 1975 follow-up date no cases of leukemia had been observed. Arguably a sufficient latent period hrr not ban allowed, and the quantification of txposur~may not bc accurate. but it is possible that here is a group of individuals from the same cthnic/gcographical background for which thcse extrapolations do not apply. The sccond group is taken from the R m k y follow-upto the Infante study and is composed of all those workers referred to as group 2 in the study who first had exposure to benzene dunng the period 1950-1960. 4 In carrying out an exercise of this 3011, sevetal USumpcioOI hve tobe rmdc to t u t tbt attrapdation since no deuit ue given of thc duntion of a p m t ia benaenaarpossd mp~thF~d-. lowing t b c u r ~ ' r p P r 0 r c hin the fisk d* the best rod wurst case hvc been adopted and an presented Wow. They aim to give a feel for the typical range of deaths that might be anticipated from this group. Corresponding to a dose of 415 ppm years asso- ciated with the observed SMR of 2100 in group 1 we obtained a range of deaths equivalent to (1.3, 12.1). The lower limit was derived assuming that all 258 workers worked for periods of lcss than one year (six months), whle the upper limit had three implicit assumptions. The first was that workers at locations 1 and 2 had a turnover rate among short term workers equivalent to that in group 1-that is 58% less than one year. The sax~ndassumption was that the remainder of the 148 workers at location 1 had periods of employment proportional to the n u m k of 'units' in o p t i o n ; W was a simplified attempt to reflect that the size of the workforce would dimiaish as the number of operatio.@ units was reduced. Finally, that the 46 individuals at location 2 who worked for longer than one year, all worked for a period of ten y w s from 1955 to 1965 at which time it was considered the benzene exposuns altered considerably due to the Change from rubbet hydt~hlotidemttfacture to vinyl film. The same exercisc was repeated using the model coefficient corresponding to 1500 ppm years associ- ated with an SMR of 2100 which led to (-3, 3.9) as the range of deaths under the assumptions described above. By 1975 only one death had occuned against an expected figure of .46 giving an excess of .54. It can be Seen that although the figures are only rough approximations, that .54 just overlaps the (.3, 3.9) inteml md is not coatlined in the (1212.1). In this context the lomrlirait is not regwdalu bang puticukrty r e p ~ U t i v de the murl wwlt profile of thecohort,md tha is asuggation that the actual excess is not as large u mticipatcd.This my be due solely to thc assumptions made in crlcukthg the figures, that sufficient timc had not elapsed to mani- fest thtcxccss,or that the original projections were derived for individuals with greater than 5 years exposure (we have extrapolated to 6 months for the lower estimate, and used periods of less than five years for the proportionality assumption in location 1.). We can either uncritically accept the projections made by the authors, carry out relatively unproductive exercises similar to the one above to scc whether the risk is 1-es4tab,lish whether 'negative studies' exist. or concentrate effort on determining a better undastandq of the wchanu.m invdved (which might dictate peak values of expoaurt being included). Without a clearer understanding of thc mahanism, or data points established in the low dosc region so that the shape of the simplified dosc re- sponsecurve is characterized. the debate will never be satisfactorily resolved. REFERENCES 1. R. A Rias&y. R. 1. YOUIL. d A. B Srmtb, "WCIXin UJ &mcne Worken." Ammcan Jour01 o/ lrdrornd M ~ C I I U2, 217-245 (1981). 2. M.G O ~ JL, C.Towarcad, W A. Fishbcck. and R. A. Lu~grr. "Molulity Among ldlvdurls Ofnrpallodly Exposed 10 Bc- cmene." A ~ I W Suj Emwramentd flmlth 33.3-10 (1978). 3 M W I ~ . " ~ C C U ~ ~ U O ~ to ~enrrneA: R ~ W C W Cuclnogrc and Related HeJtb Effects F d h a chc U S Suprenu Caut Decas~w."A m r a n J o u n a l o/ Idnsmcrl ~M&CIM 1. 133-243 (1980) 4 N M H m r J J Thorpe. "Eptdcnuologyof Bcm.mc-induccd L&cnua" {Post hunag commcnu of the Amencan Petroleum Insutute. EPA Doc No QAQPS 79-3. Pan 11. 1981) Lutter to the Editor The Methodology of Risk Assessment William R. Gaffey' In their rccent paper on a risk assessment of leukemia mortality associated with occupational exposure to benzene, White er d.`"invite comments on their approach. Several things might be said about thdr choict Of d a nspop~ed d and their CX- posure estimatcs. However,I wish to raisc a basic question about the methodology that I believe may be more important. For practically all cancers there is a background mortality that varies with age. In such cases. the assumption that the relative risk associated with some exposure is constant across all age groups is falx by algebraic nectssity. To see that the relative risk must be age depen- dent, consider the author's Eq. (4), but denote the dose response CUNC by f(d).That equation states in effect that the relative risk is that produces a relative risk of 5 when Po =.2, it cannot produce a relative risk of 5 when Po =.5. Since the authors state that the assumption of a constant relative risk justified their use of the SMR to estimate that relative risk, and siwe the assumption cannot be true, it seems to me that their procedure is not valid. It isn't so much that the SMR my overestimate or underestimate the rdrtivc risk, as they state on page 202, it's that there is no relative risk to be estimated, since it isn't constant in the first place. At the risk of jousting with a sacred cow (and of mixing metaphors) I suggest that the real problem is with the concept of relative risk, which has a seductive appearance of svaightforwardnessuntil you look at it carefully. It may be time to rethink the way in which we charactcrize the possible hauuds associclted with chemical exposures. When d is constant the relative risk is a dccrcasing function of Po. In plain words,if you have a dose `MonsmroCompanv. Boo Nonh Linbergh Blvd.. 3.Louis. Missouri 63167. REFERENCES I M C Wtutc. P F Infante, and K. C Chu. "A Quantitative Estimate of Leukemia Mortality hsoaaicd wrh Ormpruonll E x p O s ~ rto~ k~cnc.R"isk AMIVSIS 2. 195-203 (1982). 5 Rwk.4- Vd(Nal.lW Letter to the Editor Benzene and the One-Hit Model Jerry L. R Cbandlet' Future adverse health consequences subsequent to occupational exposure to benzene were estimated from epidemiologic data by White eral. A funda- mental scientific und kgal issue is the question of the relative health risks associated with chronic "low" I d exposure t~ If f a c t - M s c i t n ~an~d value-based s o d poky an to be articulated with d t y , then the bcst estimate of the true human risk should be sought The best estimate of the true humur risk demands thc systematic evaluation of dl relevant facts which may quantitatively influence the risk estimate. In addition, the assumptions and the implications of the predictive health model must be evaluated for consistency and coherency with the known facts. White et d."selected the onehit model for ben- me, primuily bocaw it is a simple model." The onthit model is not mmiy a statistid distribution but is mathematically derived from a set of specific assumptions which relate the "dosc" to an irrevers- ible biological proctss. Mathematically, the one-hlt model describes a continuously increasing saturation function which starts at zero and aspptotically ap- proaches one with increasing dosc. Chemically, the logic of the onohrt model requires that a single unique chemical event constitutes the causal biolop cal m n t . The causal biologd event must be ini- tiated in a single all in order to be unique. In addition, the chemical went must be irreversible to be in accordance wilh the mathematical assumptions for a Poisson process. A critical determinant of the model's risk predictions is the assumption that the number of irreversible biological events is directly proportional to the dose. hence the critical public h d t h conclwion that a ample proportional relatron- '7412 Churchdl Road. McLczn. Viwma 22101 ship exists between the dose and the response at "low" doses. Is the one-hit model an accurate predictor of the future incidence of leukemia in exposed populations? White et ul. state "the current biological evidence concerning the chemical inductioa of leukemia was considered insufiicieat to defend the w of a mort complex model" Unfortunately, thc nature of the evidence considered to be insufficient was not surd The notion that "simplicity" is adequatejustification for predicting future incidence of human disease disguisesat least four substantial scientificissues: 1. The fundamental issue of whether or not the dose-nspowc relationships for btnzene an continuous functions or arc discontinuous functions is Side-stqped by tbt hpti&ity argument. From a public health paspcctin, the critical issue of the existence of risk outside the observable range is thatby avoided. 2. Justification of selection of the bar model by noting its simplicity also leacis to the notion that thc response is proportional to dose. Of the infinite set of possible relationships bctwecn dosc and response, the assumption that a simple proportional relationship exists needs to be justified. Since it is possible that the molecular mechanism of benzene-induced toxicity involves a set of complex nonlinear biochemical reactions, the simplicity argument may eventually be shown to be inconsistent with the molecular mechanism. 3. The argument of simplicity is insufficient scientific rationale to justify the strong ConcIusion that diseasc incidence is Linear in a heterogeneous human population. Hit theory is based on the premise that rare stochastic events occur as first order procases in a 7 I 8 hanogeneow population. Discussion of the weight of Cvidcnce supporting the aluismjitivity of all workers toward bcnzcne intoxiation is needed if this assumption of the modd b to be credible. 4. Metabolic considerations may be critical to understanding the functional relationship between the dose and response. Since the parent benzcne molecule is a relatively unreactive molecule at biologic pH and temperature, a spontaneous irreversible reaction with bio- molecules at a biochemically significant rate is deemed unlikely. This suggests that the presence of benzene itself in a cell represents only the distribution of a foreign substance in the body and not the critical irreversible event assumed by the model. Benzene is metabe Id to phenols and quinones by saturable enzymatic processtJ. Phenols, quinones, and unstable oxygenated benzene derivatives which lead to the formation of phenols and quinones are substantially more reactive than benzene itself. Kinetically. these reactions are not simple unimdoculu reactions but fdm hi* order, srtunble lriaetia oxygenated meubolic daiVrtivg uc suffeeatly lcrtiw corrkntly bind to tissue ~nStituCnts,including genetic elenrmls. Tbc imporunct of these Ftlctions in the mldar mcchrrnisms which determine the dose response rela- tioaships for benzeae toxicities COuLh (and should) be evaluated in estimating the risk of benzene. Simplicity in model selection is desirable when the selected model can account for the complete set of factual observations. However, our current under- standing of human disease proccsses, including ben- zene toxicities, suggests that they arc multifactorial in nature and more camplex than a single, unique, irreversibleevent. To be credible, prediction of future disease incidence netds to be consistent and coherent with the intertwined biochemical, cdlular, pbysiologi- cal, and genetic data. For thesc reasons. I believe that the best estimate of the true human risk for bcnzene wiU require a broader and decpcr evaluation of the existing sct of relevant facts.