Document rBmpZkpwN6jjLmx43yYQ3RqyG

&31 Occupational Epidemiology t } Author Richard R. Monson. M.D., Sc.D. Department of Epidemiology Harvard University School of Public Health Boston, Massachusetts I CRC Press, Inc. Boca Raton, Florida AP00016911 Library of Congress Cataloging in Publication Data Monson, Richard R Occupational epidemiology. Bibliography: p. Include! index. 1. Occupational diseases. 2. Epidemiology. 1. Title. fDNLM: 1. Epidamioiogie methods. WA950M755oJ RC964.M66 616.9*903*072 ISBN 0-8493*3793*4 79*27046 This book represents information obtained from authentic and highly regarded sources. Reprinted mate rial Is quoted with permission, and sources are Indicated. A wide variety of references are listed. Every reasonable effort has bean made to give reliable data and information, but the author and the publisher cannot assume responsibility for the validity of ail materials or for the consequences of their use. All rights reserved. This book, or any parts thereof, may not be reproduced in any form without written consent from the publisher. Direct all inquiries to CRC Press, Inc., 2000 N.W. 24th Street. Boca Raton. Florida. 33431. / 1980 by CRC Press, Inc. ^ fteerfnd Printing, 1981 International Standard Book Number 0-8493 *5793*4 Library of Congress Card Number 79*27046 Printed in the United States Epiderr. rence of <* hgg no pa: influenza. Because oi rence of ir As earl emerged.. Political t. chat data of this reoi could be c forms the There at and at leu: ject matirIn some U where thet cultures t < terminok'i epidemic My tra*. in evaluat diseases i? tation of . of occupy The idm used in c.' general rv ies of oc. used. Ea. known to lions or u crally bL>-, of drugs c occupatu The su. tional epK tier of inu have dealt and cheni illnesses. . exposure cian couk However, working v able symp hasstopp 1 have i AP00016912 AP00016913 134 CXmpMiiNuf CpiJn>fe|y There U no analogue to (he HWE to a proportional mortality analysis, since by definition the total number of deaths expected must be equal to tbe total number of dcalhi observed. However, tbe HWE probably has some influence on cause-specific proportional mortality ratlot. SMRs for all cancers tend to be higher than SMRs for all other causes, usually by 10 to 20%. Therefore, even if there is no absolute eacess of cancer among a group of deceased persons, for mcthodologic reasons an eacess of 10 to 20% might be expected. Note that in tbe realm of rate ratios, this is a very weak association. Also. SPMRs tend to be more inlcrpretable for uncommon causes of death than for common causes. If the rate of some uncommon disease such as lung cancer Is increased by. say, 200%, the SMR and the SPMR will each reflect this increase. However, if a common illness such as circulatory disease is increased by 200%, the SMR will still reflect this increase, while tbe SPMR will be increased considerably less. Also, whereas SMRs for two or more causes are essentially independent of each other, SPMRs for two or more causes are interdependent, since the sum of tbe expected numbers must equal the sum of the observed numbers. In spite of these limitations to SPMR analyses, they are most useful and provide information that usually differs iiltle from that obtained ia an SMR analysis. It is especially instructive to do SPMR and SMR analyses on the same data, so as lo eval uate interpretation of the mulls of these two methods of analysis. B. Rv--pta -- Rubber Workers Consider again the data in Tabled.!. In assessing whether a specific cause of death is in excess, one might look only at causes for which the SMR is greater than 100. If this criterion is used, only leukemia and cancers of tbe stomach. Urge intestine, and bladder are in excess among the entire group of rubber workers. However, it might be argued that because of the HWE (or because of some other bias), SMRs greater than the ``ilHsuKt" SMR should be considered to be ia excess, intuitively, one astestes such excess by dividing each cause-specific SMR by the ali-causes SMR to obtain p "corrected" SMR <CSMR). The retuilr of such a division are illustrated in Table 6.17. in Column I are the mow SMRs as in Table 6.1 (note that the SMRs diHcr slightly because the analyses were carried out at different tiaei). In Column ll are tbe CSMRs. CSMRs less than 100 are seen only for disease* other (ban cancer or cardiovascular disease. In Column III arc the SPMRs from a proportional mortality analysis. By definition tbe all-cause CSMR and tbe SPMR arc 1.00. It can be seen that tbe CSMR and the SPMR for tbe several causes illustrated are similar. Part of the similarity do doubt is due to the relatively weak association* be tween disease and working in tbe rubber industry. However, under any of the three methods used to compute expected numbers, one would wish to investigate further death from bladder cancer and the lymphatic and hematopoietic cancers. There is one tendency in Table 6.17 that may be of interest, for diseases of old persons, e.g.. bladder cancer and stroke, the CSMR is higher than the SPMR. For diseases of young persons, e.g., Main cancer aod external causes of death, the SPMR is higher than the CSMR. Whether this teadeocy is a general property of the CSMR and the SPMR is unknown. If it is a general property, it no doubt results from the way the weights differ between a SMR computation and SPMR compaiatioo, i.e., penoa-yenrs vs. numbers of deaths. C. Raample--Vinyl Chloride Workers la 1974, three cases of mpomicom of tbe Ever among vinyl chloride workers were ns Tabled.16 STANDARDIZED MORTALITY RATIOS (AND OBSERVED DEATHS) ACCORDING TO PLANT DIVISION IN WHICH EMPLOYEES EVER AND USUALLY WORKED* AH ciuitl AN turai OirlMM "kudorr CSnniril Aerospace Tim IndrrurlsJ product* frnkn abber reclaim f(SMHH| fatal * tivat _ 74 (Mil 74PIII %(**) 14(1400) aiona 7t(JIJ| 7f(SJ3) Und 1 OI7) 74 (SI) *1 (2*2) St (IMS) SI (7M> S4(IS3*) (41> 17 (330) - S3 (9079) Em _ it ob (I3J) WO*7> *4 (MS) 40 (act) 44(41) 4*077) _ Und 44 (44) S0<I3) 1405) JOSS) 44(154) 44 (JOS) 100(22) 111(1171 44(9*0) " Data dsTfcr siashJb from lhamprswtod In Table *, I. Table 6.17 NUMBER OF DEATHS OBSERVED AMONG RUBBER WORKERS AND RATIOS OF OBSERVED TO EXPECTED ACCORDING TO THREE METHODS Cause efdsaik Observed Seeds Otoerved/especred* ---------------------------------- 1 II Ul AHCWKS AN nurccr Digealvc ladder Brain Lymphatic sad hematopoietic CNSmualat CirruUtnry Ellerad AH aster causes *9*29M 344 4* MS 4*3 2449 271 7S4 0.12 0.44 a.4S 1.22 0.S0 1.0 1.00 1.17 1.22 I.S2 1.00 1.42 1.00 1.17 1.21 1.43 I.M 1-44 0.91 0.13 M2 0.49 M3 1.03 0.77 0.74 urr 1.01 O.OS a.77 Esddct7 4c*ih>i>f nakMn ewk. *1 Expected number! boxed on 5>year tft-tuBt specific nonnliij me* for u.& whin mates tSMIU toot. U kmwwdtsUnsmtSnvnbmndmdtdUye.SHaH cnnms SMa/100) sad by 0.41 to coma to 47 dtall* tt ufcmwn cause (CSMR/ID0). Ul Expccscd milkers band on t-yese ateiUre speeHk Fsapostienal monsluy rmios (or U.S. <MK mslee ctPMa/tw). * reported-** Later ia the same year, the results of two epidemiologic studies were re ported. In one. a standardised proportional mortality analysis was conducted of the fuses of death among 161 deceased persons, most of whom had worked at the same plant as the three with angiosarcoma.* la the second, a standardized mortality ratio IM Oea.-arwmeat fjmlmttnAmj 117 Tsble6.il OBSERVED AMD EXPECTED DEATHS AMONQ VINYL CHLORIDE WORKERS IN A PROPORTIONAL MORTALITY STUDY* Caatcaf death Observed Expected SPMR* Al CHKI Alcaacet DipcMivc Liver and Nbary Laa( Inti Other CNS vascular Circulatory Esuraai Afl uhar causes III IS 1.0 too 41 27.* in 1) .) ISO a A7 1100 il 7.9 1*0 s 1.2 20 i ms W as ao H U.i N 11 M.S n M ,5 7* Table 6.19 OBSERVED AND EXPECTED DEATHS AMONG VINYL CHLORIDEWORKERS IN A MORTALITY STUDY* Cause ol death OUcrved Expected SMR AH causes ABcaaccr Puerave Liver Raparaiary Brairt Other CNS vascular Cheatstco Eatereat All oiher causes 2 4*7.J 75 79 77.2 toe 19 21.7 14 7 1.7 412 75 1). 10) 7 1.1 2W 10 9.1 l 11 24.) n 142 I74S 74 S2 91.7 U 93.1 44 Eapeeted Miatbcn huiirl on iwfiinil raoteahiy raIto* lot United Stales wfcae mIh SPMR - Suadardtrad properUodal marfHj ratio ~ * Eapecied aurabos based <m aonaiuy rates for Uuiet Sum uOS> raatra. MOnebaervcd/cxpecKd. Table 6.20 analyst* conducted on the mortality ainoni DM men who bad worked with vinyl chloride, including those at the same plant aa above.*1 Among these BM4 workers were 152 who had died, including most of the 161 included in the above report. In Table 6.11 are presented the result! of the SPMR study. A S0% excess of cancer was seen, which was due entirely to cancers of Ike liver, the lung, and the brain. In OBSERVED AND EXPECTED DEATHS PROM ALL CANCERS AMONG VINYL CHLORJDB WORKERS IN A PROPORTIONAL MORTALITY STUDY ACCORDING TO AGE AND YEAR OF DEATH Table 6.16 are presented the results of the SMR study. As compared to the general population, there was essentially ao excess of cancer. However, the CSMR was <102/ Chuacmistic Caseory Observed Eapected SPMR 7J> u 100 or 1)6. There were large excesses of cancers of the liver and of the brain. For respiratory cancer, while thcSMR was dose to 100, the CSMR was 140. Thus the two studies reported similar results, even though one was smaller and based Aac el death < SO--)9 >40 II 7.4 140 17 S.1 210 12 17.2 no simply on a SPMR analysis. In this study an analysis according to age and year of death of persons with cancer was also done <Table 6.20). As can be seen, while there is no trend with age at death, there is a steady increase in the SPMR with year of death. This suggest* It*** many more persons will die from cancer because of exposure Year atdeath CI945 1905--04 1970 12 11.4 no to 7.1 140 1* 9.2 210 to vinyl chloride. lung cancer. Exposure must be reduced because of its highly likely role la causing The judgment as to whether the excesses of specific causes of death are likely to be causal is based on reason rather than any statistical significance. There can be little angiosarcoma. If some lung and brain cancer arc also prevented, so much the better. doubt that vinyl chloride (or some closely related compound) is responsible for the IV. CASE-CONTROL STUDIES WITHIN A COHORT angiosarcoma. This disease was extremely rare before il was discovered among vinyl chloride workers. The rate ratio among vinyl chloride workers must be in the lOOOi. On occasion, personnel records may be available for a cohort of employed persons, The disease occurred mainly in the men who worked for many years (10+) cleaning but it may be too expensive to review all of the records to determine the work history the reactors; these men accumulated a high dosage of exposure to vinyl chloride. While of each person, Some sample of the records is selected to be reviewed. Since a primary aagioMfcoma association is nt the top of the highly likely to be causal group. This interest in such review It to determine association between work and disease, priority nothing in epidemiology (or any science) is ever absolutely certain, the vinyl-chloride- is given to reviewing the work histories of persons with one or more diseases of interest. angiosarcoma association is at ihe top of the highly likely to be casual group. This For purposes of comparison, the work histories of "controls" are also determined. was so after the first three cases of angiosarcoma were Identified. With respect to cancers of the brain and of the lung, a causal interpretation of the A. Procedure association Is not possible at the moment. Both are relatively common cancers and are la this discussion, it is assumed that mortality data for the entire cohort are avail based on much smaller rale ratios. The cause of most of lung cancer is known, and able. Initial analyses proceed as described in Section II of this chapter, including Tables the excess seen here could conceivably be due to cigarette looking. The of most 6.1 to 6.11. If one or more causes of death are found to be m excess and If there la > of brain cancer is aot known. It is mainly of scientific interest whether vinyl chloride exposure leads to brain or 110 Ou-spMKMMf CpnieutMotr U. Trr.Mii1iril Tnufsnffrm--'---- ^-------- -- Thn-- m, e. T.i MiySea, A. C..Bd*..Mar*wH*ll.Hw Voik. 19*1. II. Mm. 0., Sbmb*. S., sol HUAS.O. S., Cwwcwurel wrwUUaeeal HriMaatMiWaUdlt in nnt-if--irr---- ta p-'-*--*-*-r^n--' P.. Shapiro. S..Stow, D.. udTo^Mi.O.Eds., etanSw.AjusWrdsW. I*??. *. 14. morn,. D,, Skafira, S.. ItMbai. U. Kmf-aa. 0. W.. HwU. S- C. bkl, A. C,, Sw*r. P. . md `llTT-'nn-.^ S.. gciwiou f Ufumc mwfcuw o mIkcum iwM worn*. N. BmgL J. U?j. If?*. isi Chapter 9 CURRENT PROBLEMS IN OCCUPATIONAL EPIDEMIOLOGY I. INTRODUCTION The purpose of occupational epidemiology it to provide information on human health that can be used to prevent human illness. At time gees by data accumulate on the relationship between occupational exposures and human diseases. Some of these data can be interpreted unequivocally and can provide a firm basis for a change in behavior. Other data are subject to more than one interpretation or may be challenged as to veracity. As indicated throughout these pages, data by themselves provide no indication as to whether or not they reflect true processes in human populations. In (his chapter a variety of problems related to occupational exposures are consulered. These problem* are illustrated by referring It recent literature in occupational epidemiology. The intent is lo provide an overview of general problems rather than to provide a detailed review of specific issue*. Excellent reviews of the known and sue* pccted effects of exposure to substances used in industry can be found in the Criteria Documents published by the National Institute for Occupational Safety and Health (NIOSH) and in the reports of the International Agency for Research In Cancer (I ARC).** For each of the studies reviewed in this chapter, the collodion, analysts, and inter pretation of (be data arc presented and the meaning of the results is discussed. Some of the studies are models on bow to conduct research; some of the studies illustrate problems In methodology that should be avoided. All of the studies illustiate the real ities encountered in conducting research in occupational epidemiology and in inter preting the results of that research. 11. OCCUPATION AND REPRODUCTIVE EFFECTS Exposure to substances used in industry nuty lead lo effects on the reproductive system in men or in women or may lead io an effect on the unborn fetus. Few data have been published on the general relationship between occupation and problems of reproduction. Until recently few pregnant women have been exposed to industrial en vironments. Any effects on male workers are difficult to detect. Any effecss on fetuses have not been studied to any extent. A. Vinyl Chloride and Fetal Deaths In October 1974, interviews were conducted with male workers exposed to vinyl chloride monomer (VCM), polyvinyl-chloride (PVC) or rubber.* Questions wereasked about the interviewee's health as well as about the outcomes of pregnancy of the work ers* wives. The fetal death rate subsequent to exposure for VCM was higher among the families of exposed men than among the families of controls (Table 9.1). This study is difficult to classify. It is not a retrospective cohort study, because t workers were no! followed either hi retrospect or in prospect for the occurrence of fetal deaths. It is not a case-control study, because subject selection was not based on presence or absence of fetal deaths. It is best thought of as a cross-scclkMial study where exposure is presence or absence of a history of VCM exposure and is presence or absence of having had fetal death. A potential problem is that there may have been a differential rate of termination of employment according to exposure- * disease status. The data seen in Table 9.1 could result if exposure to VCM did not AP00016915 Ml Oix-xpx/kwuUEpf<irm*j4ofy Table 9.1 FETAL DEATH RATBS AMONG FAMILIES OF MEN EXPOSEDTO VINYL CHLORIDE MONOMER AND AMONO FAMILIES OF CONTROLS On* FstsIdnsbraM' Mkmcvwn SHkMe*M Bnpwin (Ml gwihnthif Crude Suadsrdtwd Exposed 10. | i.l 1*3 111 CoatroU M i.l I.l I.l Ntussbcsol (csaidcwktpcs lOOpngiiMCKt. bad to increased fetal death, but if controls with foul deaths tcnninated at a higher rate ihaa opoicd workers ilb fetal deaths. This can be (bought of as selection bias. Selectton bias also was possible because not all workers eligible (or ibe study were interviewed. The data in Table 9.1 could have resulted if exposed workers wish a his tory of fetal deaths selectively agreed to participate in the study. Beeaute the interviews were not conducted blindly, observation bias was possible. The data in Table 9.1 could result if workers wish exposure to VCM over-reported, or il controls under-reported, the occurrence of fetal deaths. No information on smoking of mothers it presented. Smoking it known to be asso ciated with an increased fetal death rate.* The data in Table 9.1 could have resulted from confounding by samSrinj if (be wives of VCM workers smoked more than the wives of controls. The primary procedure used In the analysis wss direct standardization of rates, be cause the age distribution differed between exposed and control groups. Because the standard used was the sge-distribwiion of Ibe controls, the crude and standardized rales for controls in Table 9.1 were identical. For Ibe exposed, there was a substantial reduction in she standardized rate prior to exposure. Tbit occurred because the exposed group was older than the control group, nod because the rate of fetal deaths increased with ape (Table 9.2).* In this example the standardization behaved well, since the age-standardized rales as wdl as the age-specific rales showed essentially no differences in fetal death ram between the exposed and the control groups. However, the potential difficulties of direct standardization with respect to small numbers can be seen, if there had been more controls in the <20 category, and if there had been two or three fetal deaths among the seven exposed persons in the <30 category, that stratum would have con tributed heavily to a high standardized rale among the exposed group. In such a situ ation, it is best to Itaiii the analysis to overlapping age-strata io which there are suffi cient numbers, e.g., ages 20 to 29. In any standardization procedure attention mint be given to the elements being standardized. The meaning of this single study b unclear. Some of the criticisms raised above must be viewed as far-fetched. The likelihood of selection bias seems remote. The possibility of observation bias is more real, inasmuch ns the intcrviewi were conducted shortly after Ibe carcinogenic properties of vinyl chloride were announced. If anything, the fetal death rate of I.M in controls seems low, suggestingthat they may have not been simulated sufficiently to recnR previous miscarriages in their wive*. It seems unlikely that theemoking habits of the wives of these two groups of workers would differ; evea as Table 9J FETAL DEATH RATES PRIOR TO EXPOSURE AMONG FAMILIES OF MEN EXPOSED TO VINYLCHLORIDE MONOMBR AND AMONO CONTROLS ACCORDING TO AGH OFFATHER Amaffwhar Nuahtrt(pt|iiuiekt OtMp Expand Central <30 7 SI SO-M 44 SO 3S-8 ss 34 M-->4 n >35 14 4 Al MS m Ecu] 4cMh rate* Exposed Cnud 0.0 S.3 4.J SO IZ.3 M.3 IS.} M.7 7.1 M.l 4.4 at * Nwabcrof fctaidcwktpcr 100 pmaaarte*. if wives of exposed men smoked more, tbc association between smoking and fetal deaths is weaker than the association seen in this study. As suggested by the above paragraph ihe meaning of the results of this study is a matter of opinion, not of science. Different persons will rake different criticisms or different defenses. Cerufady the results of thb one ssudy are not definitive. With respect to future public health policy, it not a ma|oc concern whether vinyl chloride exposure to men tends to locicased fetal deaths among their offspriag. The association between VCM and angiosarcoma is many times stronger and therefore VCM exposure must be reduced. Such reduction would be expected to miaimizc any adverse effects of VCM on the reproductive system. However, it is of public health concern that VCM might affect pregnant women directly, la order to address ibis concern, studies must be made of pregnant women exposed io VCM and of their children. B. Paternal Occupation and Childhood Cancer I. Children so Quebec "In reviewing, for other purposes, a small sample of birth nsid death certificates of Quebec children, the impression emerged of a large number of fathers in petrol-related occupations when the cause of death was cancer."* To quantify this impresaioo, the occupation of the father was obtained from the birth certificate of )M children under the age of 5 who died from cancer. For comparison purposes, two matched controls per case were selected -- the preceding and succeeding child in the birth certificate registry. As seen in Table 9.3, approximately twice as many cases bad father* with potential exposure to petrol and other bydrocarboas. The odds ratio is (71 * 697) + (7J*31S| 2.1. The 9SW confidence interval of the odds ratio Is 1.3 to 3.0. Selection bias Is an unlikely explanation since paternal occupation was not ascer tained uatil after case* were defined. Observation bias could have led to these resells. The authors admit to a prior notion that petro-rdated occupations were related to childhood cancer. In classifying a father's occupation, it is possible that occupations difficult io classify were coded as petrol-related for cases but not for controls. How ever. the authors state that occupation was coded by a person who was Minded as to case-control status. Thus, few this reason, observation bias was not possible. Confounding by some factor related to occupation and childhood caaxr could al ways be aa explanation. However, no data on potential ronfounriiag factors were available from the birth certificate, except for age. The mean paternal age for both cases and controls was 31.4. Further, little is known as to determinant* of childhood cancer. IU CkxupttkiailCptdemblatr Table 9.3 OCCUPATION OP FATHERAT TIME OF BIRTH AMONG CHILDREN DYING OF CANCER IN QUEBECAND AMONG CONTROL CHILDREN OocupMka a# fstlier Mow-' ctucif aiitiiili Mtrhiain Other hydrprarboacipoiHic Other CUkLrcawtth esaerr a M n m Count mtldna 27 a i* *97 Total MS 771 OthhiMk - u The data can be critRzcd if they include the cases upon whom the initial impression was based. It nay be that the initial impression was correct, but waa based oo a sample of children with cancer who were simply atypical because of cbaoce. la order lo teat the correctness of the impression, independent data should be evaluated. It is not stated in this report whether or not this was done. In the analysis of data a large number of occupations were coded. The data in Table 9.3 are derived from a much larger Table. The association with hydrocarbon-related occupations is partiaQy a result of looking through a number of occupations and grouping those wMcb appear to be related to hydrocarbon exposure. However, this grouping seems lo have been done without regard for disease status; for example, among the "other hydrocarbon exposure" grouping were three cleaners among com urols and only one among cases. A criticism might be made that n matched analysis was not done. However, if the controls were matched to cites for purposes of control of Confounding, the lack of n marched analysis would only dilute any association between exposure and disease. The occurrence of an odds ratio of 2.1 (instead of 1.0) could not be attributed to the nonmatchcd nature of the analysis. Further, the matching was done essentially for pur poses of convenience rather than of control of confounding. A matched analysts was not called for. It is difficult to interpret the results of a stogie study such as this. While the sssociation is probably real in this population at this time, it is not necessarily causal. The fact that the association is statistically significant does not mean that it could not have occurred by chance. The only change in behavior dictated by this study should be to collect further data. 2. Children in Finland In response to the above study, an analysis was done of data collected in a Finaisti study of childhood cancer. Cases were children under the age of 15 who developed, rather than died from, cancer. One coatroi for each case was selected; the control was the child born immediately before the case in the same maternity welfare district. Pa ternal occupation was routinely recorded for all children at the first visit of the mother lo the maternity welfare center. As seen in Table 9.4. there was no association of paternal hydrocarbon exposure with childhood cancer. The 95% confidence limits of theodds ratio of 1.0 were O.g and 1.3. The data io this study indicate no association between paternal hydrocarbon expo sure and childhood cancer. Criticisms must be directed at factors that may have led to the failure to detect a true association, should one have existed. Table 9.4 OCCUPATION OF FATHER AT TIME OF CONCEPTION AMONG CHILDREN DEVELOPING CANCER IN FINLAND AND AMONG CONTROL CHILDREN tyrupwOw at futhw Chitdrca with cancer Coatrols its HinVcciiWnnpcuri Other IM IO* M3 743 Tout Oddi ratio - l.e 2 IU Random misdassification must be considered. With respect to disease, liule mUctassificaiioa Is Ukdy. Few children with the diagnosis of cancer do not have the disease; few children without the diagnosis have latent disease. However, the determination of petrol exposure in the father is not a precise procedure. It is possible that some fathers said not to have exposure in fact did and vice versa. However, in order for an odds ratio of 1.9 lo result from random ntisciatwfkation, there would have to be ao rela tionship between true paternal occupation and classified paternal occupation. This seems unRfcdy. Selection but could have led to these results, assuming that paternal exposure to hydrocarbons led to childhood cancer, only through a rather convoluted process. Cases who toad fathers wjtb hydrocarbon exposure would have had to be selectively excluded from the study or controls who had fathers with hydrocarbon exposure would have had to he selectively included. Inasmuch as she data were collected prior (o any hy pothesis at to hydrocarbon exposure of childhood cancer, this possibility seems re mote. For the same reason, there seesns lo be little likelihood that paternal occupation was ascertained differently from cases and controls. In order for confounding bias to have led to data in which there was no association between exposure and disease, negative confounding must have occurred. Some factor that was a cause of childhood cancer would have had to be more common among the controls than among the cases. Since little b known as to the causes of childhood cancer, such a factor would be difficult to identify. There is known to be a weak association between prenatal X-ray and childhood cancer.1 If (here were more X-rayed children among fathers with no hydrocarbon exposure, negative confounding would be pocsible. However, the association between childhood cancer and prenatal X-ray is very weak (OR ** 1.5) and could not mask a trueodds ratio of 2.1. 3. Comment These two studies led to data that are not compatible on statistical grounds. The 95% confidence limits do not overlap and thus it is quite unlikely that results such as these would result simply because of random variation in the selection of two samples from (he same universe. However, the data from these studies are not at all atypical of data collected in the epidemiologic setting. It is relatively common for two studies to lead to grossly different results that do not seem to be explainable for raethodoiogk reasons. Whosever an association is first reported, as in the Quebec study, it is reasonable lo report (be association and to comment briefly on the possible meaning. If there is broad general troth to the asaociatfoo. this wW be quickly recognized and substantial- AP00016918 ' tr jt ,' 1M CtecupiMnii CpjJfrmotvgf lif dat* will soon be a&sembled. There is no need for the authors either lo defend their daLx against detractors or to announce (heir wide genesaluabilhy. The author* ol the Quebec study behaved in a proper conservative manner by recommending only that Imther investigation be done. When ihe results in Quebec became known, persons with interest in the general prob lem considered bow to examine further the relationship between paternal hydrocarbon exposure ami childhood cancer. Among these persons were there who bebered that the association was casual, those who befieved that the association was ooosease, and those who were agnostic. Each type of person considered Ihe possibility of collecting further data: some may have wanted to collect confitmatory data, some naay have wanted to collect contradictory data, and some warned simply to collect data. There Is do necessary connection between a person's opinion as to the truth of some associ ation and his desire to collect data that confirms or refuses the association. However, it might be supposed (hat there will be a tendency for true believers to collect contra dictory data. Agnostics will not necessarilycollect Irae data, for their teeming neutral ity may only be a mask covering some hidden bias. The first data to appear after Ibe initial report of some association tend to be tbe result of expedience rather than of a desve to confirm or refute tbc association. The Finland paper resulted from an analysis of data collected for another reason and pre sumably free of bias. Had this paper been authored by representatives of the oil indus try, there would be a natural reaction that the industry was out to cover-up the harmful effects of paternal occupation- However, if a confirmatory paper had been published by representatives of the ol industry there would be a strong desire on tbe pan of the general scientific community to believe the causal nature of the association. As time passes, other data will appear on the relationshipbetween paternal exposure to hydrocarbons aad childhood cancer. One might expect that data that show no as sociation will be viewed as beiag "uninteresting" and will not be published, while data that show an association will be published. These are emotional human responses to data. However, it seems unlikely that these is a strong tendency in science (or such preferential publication of data that show association. As in this situation, whenever a positive study appears, a negative study frequently follows. The judgment as to tbe ultimate truth as to whether the exposure causes the disease is not a matter of adding up the positive and tbe negative studies. Each study must be considered as to tu merits. It is possible that aM studies are correct; the exposure may cause the disease in one population but not mothers. It is bos necessary that universal truth be the final result of the consideration of tbe results of a number of scientific studies. With respect to ihe specific association between paternal hydrocarbon exposure and childhood cancer, the question is still undecided. There is no obvious reason to fault one or the other study. While on general principles it would be wise to limit exposure to hydrocarbons, there is as yet no need to set a specific standard of exposure because of likely damage to the future children of a nsn ^ to petrol. C. Summary Much work is needed in general on the etiology of birth defects and specifically on the possible role of occupation. I believe that epidemiologic surveillance systems on industrial populations are ideally suited to such study. The outcome of birth occurs close in lime lo the exposure; a number of medical visits and insurance payments take place during a pregnancy and delivery; any abnormality of the child usually is apparent at or shortly after birth. Bptdcnaotogk surveillance systems on industrial populations should be designed so in as to incorporate data oa prcgnaocies among wives of male workers. Ai the preem ployment history, information may be obtained as to previous reproductive history. If pregnancy expenses are paid by a company-supported insurance program, data on ihe pregnancy and its outcome should be readily available from the insurance company and/or the hospital. Because any adverse outcomes of pregnancy rdated to work are likely to be of rela tively low frequency, large amounts of data will be necessary to detect any association. Therefore, it scents reasonable that groups of companies within specific industries pool their data or share their findings. Much of the data collected will be routine and of little interpretive value. However, if an abnormality of pregnancy is caused by tome parental work exposure, it should be detected as soon as possible. Elimination of tbe responsible substance has the potential to lead in a relatively short time to elimination of the adverse health effect. This contrails with the years or decades needed to dem onstrate the successful elimination of an occupational carcinogen. IU. OCCUPATION AND RESPIRATORY CANCER Currently, there ucooccrn over the role of chemicals in the etiology of cancer. It is hypothesized that long-term knv-levei exposure to environmental pollutants may be responsible for a high percentage of cancer.* However, u is diffcislt to study this ques tion in the general population; no records exist on past exposures to these pollutants and doing proapcciivc cohort studies would be expensive and of questionable feasibil ity. Persons working in industry are exposed lo higher levels of many chemicals; some rough estimate can usually be obtained as lo past exposure; the conduct of retrospec tive cohort studies on industrial populations is relatively inexpensive and feasible. Therefore, muds can be learned about the role of chemicals in the development of human cancer by studying industrial populations. There is a wide spectrum of current knowledge as to the relationship between indus trial exposure to chemicals and cancer. For some substances there is an unquestioned relationship between a substance and a cancer; standards nf exposure to the substance exist, are enforced and seem to be effective. For some industrial sellings there is agree ment that the general exposure is harmful even though there is no specific association bciwcca a substance and a disease. Standards In the setting are aimed at general envi ronment rather than specific exposure. In other settings there it a contested association between exposure and disease; to some observers an association represents cause and effect while to others the association is an example of bias. Finally, there are situations in which the data arc conflicting and most agree that no conclusions are possible at the moment. In this section the relationship between occupation and respiratory cancer is consid ered. For some occupational exposures, there is a clear-cut causal association with respiratory cancer. For other exposures, excess respiratory cancer has been reported but disagreement exists as to whether tbe association is causal. A. Nickel and Reapiratoey Cancer in 1949 on the basis of clinical reports, cancers of Ibe none or lung among workers exposed to nickel were classified as industrial disease*.*" To quantify thiMssociatton, a study was done of the occupation of 15,247 men who died from brag cancer, from ---i eamxr. or from other causes." This study was done of deaths between 1934 and 1954 in an area of Great Britain where a nickel refinery was located. Tbe data were collected by a review of death certificates. Cause of death was coded in accordance whb standard procedures; tbe occupation of each decedent was that 6169 000dVl l I j , 1 i i in OccupwiooW Epidcmtohir listed on the death certificate. The relationship between cause of death and occupation is given in Table 9.$. Excesses of nasal cancer and of lung cancer among decedents in the nickd industry are apparent. It is difficult to construct a realistic argument (hat the associations seen in Table 9.S are the result of bias. AN deaths of men residing in ibis area were included; therefore there was no selection. Selection bias could have resulted if men who were employed in the nickel industry and who died from "other causes" migrated from Great Britain before death, but this seems unlikely. There is also no reason to believe that nasal and lung cancers were overcertified as the cause of death in men who.worked in the nickel industry. Confounding by cigareue smoking could be considered for the excess of lung cancer, but is extremely unlikely. In those years a very high proportion of British males smoked; therefore even if nickd workers smoked more than other workers, the amount of confounding possible was minimal. Cigarette smoking has little, if any, effect on nasal cancer. A question can be raised as to the bask study design and the way the data were analyzed. The data were based solely on death certificates. No fodow-up was made of a group of nickel workers and no mortality rales were available. The study was essen tially a proportional mortality study in which the percentage of nasal and lung cancers were compared among various groups of workers. Crudely, the proportion of nasal cancer among nickel workers waa0.06S |13/(13 + 48 * 139)1. Among "other'' employees the proportion was 0.00064 and among aN employees the proportion was 0.0016, The ratio of the proportion of cancer among nickd workers io the proportion among "other" employees was 100:1. If the ratio is computed relative to the "total1' group, it is 40:1. Irrespective of which group is used for comparison, the excess nasal cancer among nickel workers is apparent. The pro pound of lung cancer among nkfcei workers (0.24) was also subaamially higher than , the proportion among "other" workers (0.05). Because of the strong association between employment in the nickd industries and cancers of the nasal sinuses and lung, the study design and analysis could be simple. The error introduced by the use of proportional mortality is perhaps 10 to 304k; the error introduced from the lack of age-standardization is also small. These errors can have little effect on the 40004b excess of nasal cancer or the S00% excess of lung cancer among nickel workers. The causal nature of the association can readily be accepted, However, the determination of a safe levd of exposure to workers in nickd refineries is much more diffkuli. Safety caanot be considered to be achieved until no excess nasal or lung cancers occur. For nasal cancer, because of its rarity, this means no cases must occur among akket workers. Per lung cancer, which is largely determined by cigarette smoking habits, some nonzero number must be expected among any group of persons who smoke. In 1970, the occurrence of deaths among a cohort of nickel workers was reported." Eight hundred and forty-five men who had been employed in a nkkel refinery in South Wales were followed from 1939 through 1966. Excess nasal cancer among nickd work ers was suspected io 1927 and the refining process was changed between 1923 and 1933."-** Therefore. H was of particular interest to determine whether any excess nasal or lung cancer had occurred la men first exposed after 1923. Because this was a retrospective cohort study, the observed number of deaths could be compared to the number expected based on national death rates. The results are presented in Tnbte 9.6. No nasal cancer occurred la men who were first employed after 1924; only a slight excess of htng cancer was seen in these men. It might be concluded that ibe change in process led to the prevention of nickel'induced cancer, However, in 1977 an updated folow-up was presented (Table 9.7). '* Among men Table 9.5 CAUSE OF DEATH AMONG PERSONS EMPLOYED IN NICKEL AND IN OTHBR INDUSTRIES IN OREAT BRITAIN leeielry or BnUpywx Nickel Ortief BetMug Suet CtMlHaiMi Other Total Will micur 11 1 1 2 25 Cuieetinib LuaSCMc* 4t 54 121 7) SO] 799 Other cauecs 1)9 MS 2.MJ 2,729 1,994 14.423 Tablc9.6 OBSERVED/EXPECTED CAUSES OF DEATH AMONO NICKEL WORKERS ACCORDING TO YEAR OF FIRST EMPLOYMENT; FOLLOW-UP 1939 to 1966 Vmraitimfpinyai Hateicaacer <IS2S 1921-29 193a--44 Mro.tr 9/0.01 are.u Cuwcofdeath 1*5/119 4/2.3 4/J.S Otherraaiei 241/201/7 21/2J.7 44/13.9 II* who started employment between I92S and 1929, one case of nasal cancer occurred. An excess of lung cancer was seen in both groups of men who were first employed after 1923. While this excess is relatively much smaller than that observed in men who started working before 1925, it suggests that some workers employed after 1924 devel oped hiag cancer because of their work. * in this example, data on men who started working between 1923 and 1929 are being evaluated in 1977 to determine whether or not nkkel workers develop exems cancer. It might be assumed that the risks to workers of today will be no higher than the risks to the workers in 1925. However, it seems clear that one cannot conclude that Use changes instituted in nkkel refining around 1925 were sufficknt to prevent the devel opment of work-related nasal or lung cancer. The specific agent responsible for the excess cancers is not known. While both nickel and nickel carbonyl gas are carcinogenic," excess nasal and lung cancer has also been observed in men exposed mainly to nkkel ore dusts." Had control* been instituted in 1925 only against elemental nkkel and nkkel carbonyl gas, the reduction in stmt respiratory cancer might have been much less. A parallel to the nickel cancers may be seen in vinyl chloride. The excess occurrence of a rare cancer (angiosarcoma of the liver) was recognized on dinkal grouqds.** Clin ical recognition was possible because the disease had few other causes andhad very high rale ratio associated with the exposure (100+ ). The process of producing vinyl chloride has been changed so that exposure is greatly reduced.** However, it is un known as to whether these controls will prevent the occurrence of angiosarcoma. Also, the possible excess occurrence of cancers of the lung and of the brain among HQ OmapuiaaTapidcmie*w Tabic 9.7 OBSERVED/EXPECTED CAUSES OF DEATH AMONG NICKEL WORKERS ACCOR D1NOTO YEAR OP FIRST EMPLOYMENT; FOLLOW-UP 1934 TO 17L Cwittftek Tarar rkwplnyaiiei Mr*r roarer Logcater Odor row IW-- tns-H *4/0. n wa.oj> MS) I3*/I14 VIA Vi.i 5*4/371.0 n/xej 64/54.9 ' DecuadewfeaedMalewKcrhMdMtroalUwttttetM. workers exposed to vinyl chloride have been reported in a proportional mortality study.11 if vinyl chloride leads to these cancers, the rate ratio is many times less than that for angiosarcoma. Further, these cancers arc much more frequent ibori the fett ers! population and might be expected to occur among vinyl chloride workers for noncausal reasons. It may be difficult to demonstrate that tbe occurrence of one of these disease* in a vinyl chloride worker is not due to vinyl chkxide exposure. Also, it should be clear that at the moment the decision as to the "safe" level of expostsre to vinyl chloride is a matter of opinion rather than of science. In one study no excess cases of angiosarcoma. lung cancer, or brain cancer were reported.** Jt b possible that workers at this plant were exposed to safe levels of vinyl chloride. How. ever, it is also possible that insufficient time has passed for any vinyl chloride-induced cancer to develop. The selling of a standard of exposure to vinyl chloride had to be based on a process of negotiation between government and industry. The standard must be conservative, i.e., because of the uncertainty associated with what is a safe level of vinyl chloride, tbe standard should, if anything, be too low. B. Coke Oven* and Respiratory Cancer* In 1962, a collaborative study of the mortality experencc of steelworkers was initi ated by the U.S. Public Health Service, tbe Graduate School of Public Health of the University of Pittsburgh, and three large steel companies.1* The purpose of the study was to assess conflicting reports of excess illness among certain groups of steelworkers. As of 19)1, these were 59,072 men employed in seven steel plants in Allegheny County. Pennsylvania. In 1962--64 an attempt was made to determine the vital status a* of 1/1/62 of each of these men. There were 32,263 men who were sill employed, 6346 who had retired, but were alive, 15,650 wbo had left employment but were alive, and4716 who were deceased. Only 97 mat could not be located. As seen in Table 9.1, the observed mortality among this group was 82% of that expected on tbe basis of Allegheny County rates. Aasong all steelworkers, there was ao overall excess of cancer. The next step was to break down the study group of ueelworkcrs according to their usual area of employment, So as to minimise the difficulties associated with persons changing jobs, an analyst* was done including a man in a given wmk area only if be worked there for at least 5 years. In order to durinate the over-esthnaiioo of expected deaths, the death rates of men who worked in each work area for at least 3 years were Tbit Kttfoa It taka from * to hopak-- ea the ovira--cM m finkooMtl Um." Set Tab6e9.t OBSERVED AMD EXPECTED DEATHS AMONG STEELWORKERS* Ca*co44ath Oharved expaad ndt1 AiCSUMt Alcoa* CHSrucehi Heaidisa* ErlenMl Alother gmms 4*1* nos MS IMS 474 963 5764.4 1441.4 464.5 2511.5 4S0.5 1449.1 *2 U 7* (2 105 64 - bpouU *okm hood oa oooahiy raw* Ur Atkghay County. Pua)4wl. * gMR Stiadnaiad aardUv row HI Table 9.9 OBSERVED AND EXPECTED DEATHS FOR ALL CAUSES AND ALL CANCERS AMONO STEELWORKERS EMPLOYED AT LEAST FIVE YEARS IN SPECIFIED WORK AREAS* All tiua Akcaacer Work 6m CokepbM Cold raductaaoiar OemralMUar MatMeeace UntiMuh Bo* wait* Non. whin Wbhe Hon- hiu Wbhe New- MK White Now U< White Non- white Oh*. 156 M 71 14 Rap. IMS 11.7 70.7 4.) Oh*. 29 41 11 1 Bap. 24.5 194 144 14 162 11.4 69 46.1 m na.4 it 141 16 16.4 14 11.4 74 70.1 1 J.S m 364.9 a 424 65 59.1 i 10.7 * Expected Humbert bead a aarulior rota tar *N iialwrttu who wetkod at teat* yean. compared to the rates of all steelworkers in the study who worked at least i years. Tbe results of this analysis are presented in Tabic 9.9. Among all causes of death, no `'healthy worker effect" is seen, since tbe mortality rale used for comparison was that of all steelworkers rather than thnt^of a general population. Moderate excesses of death are seen among oonwhile workers in tbe coke plant and the cold reducing mils and among white workers in tbe general labor and mamicnance areas. With respect to all cancer, however, a striking excess jt seen among nonwhite workers in the coke plant. Meat of this excess was found to be due to lung Several questions were raised by ibis fading: