Document zQXyKQz0ZZ5kaLDgnz7pD1ZgB

Scand J Work Environ Health 1998;24suppl2:71-80 Review and meta-analysis of studies of acrylonitrile workers by James J Collins, PhD, John F Acquavella, P h P Collins JJ, Acquavella JF. Review and meta-analysis of studies of acrylonitrile workers. Scand J Work fnviron Health 1998;24:71-80. Twenty-fiveepidemiologicstudiesof acrylonitrileworkers were reviewed and subjectedto meta-analytictechniques in thisstudy to assessthe findingsfor 10cancer sites.The analysesindicate that workers with acrylonitrileexposure have essentially nullfindingsformost cancers,includinglung [meta-relativerisk (mRR)0.9,95% confidenceinterval (95%CI)0.9-1.l],brain (mRR 1.2,95%CI0.8-1.7), andprostate(mRR1.0,95%CIO.7-1.4)cancers.Bladder cancermteswere elevated (mRR 1.8,95%CI 1.O--3.4). but theexcesswas not dose-relatedand was limited to plants with aromaticamines.Therefore,the bladdercancerexcess is unlikelyto be related to acrylonitrileexposure.Some evidenceof publicationbias was found in the examined literature,but thebias did nothaveasignificantimpactonrisk estimatesforindividualcancers.It was concludedthat the available studiesdonot supporta causalrelationbetween acrylonitrileexposureand cancer. Keytern aromaticamines, bladder cancer,brain cancer,lungcancer,prostatecancer,publication bias. Acrylonitrile (CAS number 107-13-11) is a high-volume industrialchemical used in the manufacture of acrylic fibers, resins, plastics, rubbers, and other chemicalssuch as acrylamide. Commercial production and use began in the 1940s. Worldwide over 3 billion pounds (1 360 791 metric tons) of acrylonitrilewere produced in 1994, and demand continues to increase (1). Published studies on the cancer rates of acrylonitrile workers were reviewed by Rothman in 1994 (2), but substantially more evidence has now been provided by 4 cohort studies reported in this issue ( 3 - 6 ) .The purpose of this article was to review the totality of the evidencenow available, including that provided by these 4 new studies and also unpublished studies we have been able to identify. Workplace exposure Inhalation is the primary route of acrylonitrileexposure, although exposure can also occur through dermal contact. Average inhalation exposure appears to have been highest in acrylic fiber production, especially in polymerization and spinning processes (3,4).Exposure opportunity is also significantin the production of acrylonitrile, acrylonitrile-basedresins, and nitrile rubber. Current operations have exposures at or below 0.5 ppm for an 8hour time-weighted average (TWA,,), but exposure lev- els measured by area airmonitoring were much higher in the past, exceeding a TWA,, of 20 ppm (P Stewart, National Cancer Institute, personal communication, 1996). Workers have frequently noted acrylonitrile odors in the workplace, and the odor threshold for acrylonitrile is known to vary between 1.6 and 22 ppm (5, 6). Excess cancers were found in chronic exposures of rats at 20 ppm (7 & one unpublished report "A Two-year Toxicity and Oncogenicity Study with AcrylonitrileFollowing Inhalation Exposure of Rats" by Quast et al). Methods We identified studies for this review primarily through a MEDLINE search using the key words cancer and acrylonitrile. We also searched MEDLINE using several synonyms for acrylonitrile (ie, cyanoethylene, 2-propenenitrile, vinyl cyanide, and acrylic fiber). Additional studies were identified through bibliographies (2 ,3 , 4, 6, 814),contacts with authors, and contacts with companies in acrylonitrile-relatedbusinesses. Solutia, Inc, St Louis, Missouri, United States. Monsanto, St Louis, Missouri, United States. Reprint requests to: Dr James J Collins, Solutia Inc, 10300 Olive Blvd, PO Box 66760, St Louis, MO 63166-6760, USA. [e-mail: JAMES.J.COLLINS@solutia.coml Scand J Work Environ Health 1998,vol24, suppl2 71 Studies of acrylonitrileworkers We identified29 studies.Twenty were published (1538), 4 were presented at this conference (35-38), and 5 were unpublished ("The Mortality Experience of Monsanto Workers Exposed to Acrylonitrile"by Zack, "Cohort Mortality Study of the Scotts Bluffmaton Rouge Uniroyal Plant" by Herman, "A Mortality Study of Workers Potentially Exposed to Acrylonitrile During Start-up: Monsanto, Decatur Plant" by Gaffey & Strauss, "Mortality and Cancer Incidence among Workers Exposed to Acrylonitrile at the Memphis Plant" by Burke, and "Mortality and Cancer Incidence among Workers Exposed to Acrylonitrile at the Beaumont Works" by Burke). We omitted 1 study because the authors did not present results in terms of relative risk (RR) estimates (21) and 2 studies because they concerned only workers exposed to both dimethylfomamideand acrylonitrile(27,28),while a slightly earlier study on the same work force provided data on all acrylonitrile workers (26). Our analysis strategy focused on the evaluation of heterogeneity as an indicator of factors that need to be considered in making a proper causal inferenceabout acrylonitrile and cancer. Precision issues were a subordinate concern. We evaluated heterogeneity via graphic and statistical methods as described by Greenland (39) and Dickinson & Berlin (40). We used a fixed-effects model to calculate the meta-relativerisk (mRR), a measure of the average ratio of disease rates for those with and without acrylonitrile exposure, as an inverse variance-weighted average of relative risks from the individual studies (39). We used SAS (statistical analysis system) software to do the calculations and validated our program using Greenland's data for coffee and coronary heart disease (39,41). When significant heterogeneity occurred, we calculated mRR values and related confidenceintervalsusing a random effects model which included an additional component in the variance of the relative risk of each study to reflect greater-than-expecteddifferences among studies. This additional variance component was computed from the heterogeneity chi-square statistic, as described by Shadish & Haddock (42).Greater heterogeneity tends to equalize the contributionof individual studies to the metaRFt and increases the width of the confidence interval. Calculations of mRR typically utilize a logarithmic transformation of confidence intervals to derive standard errors and inverse variance weights. This transformation could not be utilized in studies for which there were no exposed cases or in which the lower confidence limit was 0. In these instances, we set the relative risk and lower confidence limit at 0.1, which slightly increased the mRR and slightly decreased the heterogeneity estimates. We also evaluated these data excluding studies with a zero relativerisk or lower confidencelimit, and the results were similar. We considered the impact on the mRR of a number of study characteristics, including study design (cohort 72 Scand J Work Environ Health 7998,vol24, suppl2 versus case-referent), country [United States (US) versus non-United States (non-US)], and type of industry (acrylic fiber versus other). We also examined the impact of publication status and indicators of study quality. Personal or occupational confoundingfactors were not studied explicitlyby any authors and, therefore, could not be considered in our analysis. We noted the potential for confounding by occupational exposures when specific cancers were found to be elevated only in a plant or plants where a known causal factor was also present. We used methods described by Breslow & Day (43) to examine mortality trends for individual studies by exposure levels. Publication bias Publicationbias is an important validity concern in metaanalysis, the specific concern being that the publication process is selective. A related problem is the fact that authors are selective in their reporting of findings, especially for rare diseases. In situations in which reporting was selective, we contacted authors to get missing data and incorporated these data into our analyses. In some instances when expected deaths were not reported for a specific cancer, we estimated the expected number of deaths or cases on the basis of the ratio of expected numbers of cancers to total cancer from the largest studies [ie, the Wood et al study (36) for cancer incidence and the Blair et al study (35) for cancer mortality]. Outcome data The predominant focus in the available studies was on worker mortality rates. However, one employer's studies also evaluated cancer incidence data as determined from the company's insurance system. We consider the incidence data separately. Study quality There is disagreementin the literature about whether study quality should be used to weight results of meta-analyses. Arguments against this proposition are the lack of an objective measure of study quality and the possibility of aspects of study quality imparting conflicting effects on study results. Accordingly, Greenland (39) proposed an analytic focus on individual aspects of study quality (eg, percentage lost to follow-up, percentage of death certificates obtained, type of comparison population, etc). We chose this approach as preferable to one which utilizes an overall quality-score-weightingprocedure. Results Table 1provides selected details of the studies found in the search. All but 4 studies were cohort investigations, the others being 2 nested case-referent studies and 2 general population-basedcase-referent studies (23,25,29, 34). The 2 population case-referent studies were restricted to bladder cancer (34)and astrocytic brain cancer (25) and were not specificfor acrylonitrile exposure. The nested case-referentstudies were restricted to prostate and lymphatic and hematopoietic cancers and acrylonitrile exposure (23). The predominant industries represented in the cohort studies were monomer production and fiber and resin manufacture. Eight studies were subsequently updated [17, 20, 24, 26, 30, 32, & 2 unpublished reports ("The Mortality Experience of Monsanto Workers Exposed to Acrylonitrile" by Zack and "A Mortality Study of Table 1. Description of studies found in the literature search. Author Company-location Acrylonitrileuse Study design Study period Number of workers Includedin other study Keisselbachet al, 1980 (15) O'Berg, 1980(16) Bayer, Leverkusenplant, Germany DuPont, Camden plant, UnitedStates Monomer& resins Fibers Cohort mortality Cohort mortality & incidence Theiss et al, 1980 BASF, 12 plants,Germany Resins (18) Ott et al, 1980(19) Dow,4 plants, United Styrene States Zack, 1980 Monsanto,Texas City & Monomer& fibers (unpublishedreport) Decatur, UnitedStates Wemer & Carter, 6 plants, UnitedKingdom Fibers& resins 1981(20) Cohortmortality Cohort mortality copolymerization Cohort mortality Cohortmortality Herman, 1981 Uniroyal, Scotts Bluff & (unpublishedreport) BatonRouge plants, U n M States Gaffey& Strauss, Monsanto,Decaturplant, 1981(unpublished UnitedStates report) Delzell& Monson, Goodrich, Akron plant, 1982(22) UnitedStates Marsh, 1983(23) Monsanto, Springfield plant, UnitedStates D'Berg et al, 1985 DuPont, Camden plant, (24) UnitedStates Nitrilerubbers & resins Fibers Nitrilerubbers Styrene polymerization Fibers Cohort mortality Cohort mortality Cohort mortality Nestedcase-referent Cohort mortality & incidence Burke, 1985 Dupont,Memphisplant, Monomer (unpublishedreport) UnitedStates Burke, 1985 Dupont,Beaumontplant, Monomer (unpublishedreport) UnitedS t a b Thomas et al, 1987 Populationsof northern Farmers &workers in (25) NewJersey, Philadelphia productionof plastics & and Gulf coast of Louisianarubber Chen et al, 1987(26) Dupont,Waynesboro plant, Fibers UnitedStates Cohort mortality & incidence Cohort mortality & incidence Casereferent Cohort mortality& incidence Ottet al, 1989(29) 3 Union Carbidefacilities Resins Nestedcase-referent Collins et al, 1989 AmericanCyanamid, Fibers, monomer, &other Cohort mortality (30) SantaRosa & Fortier plant, UnitedStates Zhou & Wang, 1991 FushunChemical,Fushun Fibers (311 plant, China Cohort mortality Swaen et al, 1992 8 plants,The Netherlands Fibers& others Cohort mortality (32) Mastrangeloet al, Enichem-fiber, Potto Fibers Cohort mortality 1993 (33) Marghara,Italy Siemiatyckiet al, Populationof Montreal, Tailors using acrylic fiber Case-referent 1994 (34) Canada Blair et al, 1998 (35) 8 plants,UnitedStates Fibers& other Cohort mortality with case-referent Wood etal, 1998 (36) Dupont,Camden & Fibers Cohort mortality& Waynesboro plants, incidence UnitedStates Swaen et al, 1998 8 plants, The Netherlands Fibers& others Cohort mortality f37) benn& Osbome 6plants, UnitedKingdom Fibers & resins Cohort mortality 1998(38) 1950-1977 884 No 1950-1976mortality, 1345 Includedinthe 1950-1976 incidence O'Berg eta1 (24) study &the Wood et al study (36) 1955-1978 1469 No 1950-1975 loo No 1952-1977 1950-1 978 1951-1977 352 Includedinthe Blair et al study (35) 1111 Includedinthe Benn & Osbome study (38) 1077 No 1952-1 977 326 Includedin the Blair et al study (35) 1940-1 978 327 No 1949-1976 13cases, 52 referents 1950-1981 mortalit, 1345 1950-1 980 incidence 1957-1980 mortality, 700 1 9 5 6 1 9 8 3 incidence 1962-1982 mortality, 472 1962-1983 incidence 1978-1 981 27 casesand 43 referents No Included in the g)Wood et alstudy No No 1957-1981 mortality, 1083 Includedinthe 1956-1 983 incidence pWoodet al study 1940-1 978 6 cases of non-Hodgkin's lymphomaand 2 casesof leukemia 1950-1981 1774 Includedinthe Blair et al study (35) 1971-1988 1811 No 1956-1 988 1959-1 990 2842 Includedin Swaen et al update (37) 671 No 1979- 1986 1950-1 989 484casesand No 1879referents 25460 No 1947-1991 mortality, 2559 1956-1991 incidence No 1956-1 996 2842 No 1950-1991 2763 No ScandJ Work Environ Health 7998,vol24, suppl2 73 Studies of acrylonitrile workers Workers Potentially Exposed to Acrylonitrileduring Startup: Monsanto, Decatur Plant" by Gaffey & Strause)]. Of the 14 unique study cohorts, 8 were done in the United States [19,22,29,35,36, & 3 unpublished reports ("Cohort Mortality Study of the Scotts Bluff/Baton Rouge Uniroyal Plant" by Herman, "Mortality and Cancer Incidence among Workers Exposed to Acrylonitrile at the Memphis Plant" by Burke, and "Mortality and Cancer Incidence among Workers Exposed to Acrylonitrile at the Beaumont Works" by Burke)], 2 in Germany (15, IS), and 1each in the United Kingdom (38), The Netherlands (37),Italy (33), and China (31). The average duration of follow-up was 30.2 years for the cohort mortality studies and 28.6 years for the cohort incidence studies. The percentage lost in follow-up ranged from 0% to 12% in the cohort mortality studies with a mean of 4%. Loss to follow-up was not reported in the incidence studies. The percentage of death certificates not obtainedin these studies ranged from 0% to 6% with a mean of 3%. The remaining analyses of these data has been limited to the 14 unique study cohorts and the 4 case-referent studies. Table 2 summarizesthe results for the unique study populations,All-cause mortality was about 20% less than the general population rates, and the results were heterogeneous(P<o.ooOOl). Mortality from ischemic heart disease and accidents was also less than the population rates. All specific causes of cancer examined were near or below the expected levels with the single exception of bladder cancer [mRR 1.4,95% confidence interval (95% CI) 0.9-2.01. The results for all specific causes of death were homogeneous across studies with the single exception of those for colon cancer (heterogeneity P-value = 0.0062). The heterogeneity was due to the Mastrangelo et al study (33) with a standardized mortality ratio (SMR) of 10.5 (4 deaths, 95% CI 2.9-26.9) for colon cancer. The heterogeneity P-value was 0.33 for the remaining studies, and the mRR was 0.8 (95% CI 0.6-1 ,O). Results for the cancer incidence studies were similar to the results of the mortality studies. The summary cancer incidence results were near the expected levels. The prostate cancer rates (mRR 1.4,95% CI 0.8-2.6) were slightly elevated. The incidence rates from the 3 studies for aIl cancers were homogeneous. Sincethe mortality and incidencemRR values were similar for most of the cancers and there was substantially more mortality data, we limited further analyses to the mortality data,with the exception of prostate cancer. In the analyses of study characteristics, only publication bias, country of study, and the potential for nonacrylonitrile plant exposures were predictive of study results. In this report we focus on total mortality, lung cancer, prostate cancer, brain cancer, and bladder cancer. These Table 2. Meta-relative risks and 95% confidence intervals (95% CI) for the mortality studies and incidence studies by the 9th revision of the International Classification of Causes of Death. Diseases Numberof studies Observed Expected Meta-relative risk 95%CI P value heterogeneity Mortality studies All causes 13 All cancer(140-209) Stomach (151) Colon (153) Liver (155-156) Lung (162) Prostate (185) Bladder(188) Brain (191-192) Hodgkin's Disease(201) Leukemia(204-208) Non-Hodgkin'slymphoma (202-203) 12 9 9 a 12 10 10 11 8 10 9 Ischemicheartdisease(410- -414) 6 Accidents(800-949) 6 Incidencestudies All cancer (140-209) 3 Stomach (151) Colon (153) Liver (155-1 56) Lung (162) Prostate(185) Bladder(188) Brain(191-1 92) Hodgkin's Disease (201) Leukemia(204-208) 3 3 2 3 3 3 3 2 3 2769 3739.3 783 920.8 37 48.1 55 65.4 9 13.7 315 339.4 35 33.7 30 23.1 58 59.4 7 9.7 23 32.5 22 26.3 579 707.5 200 243.2 118 121.5 1 2.7 15 13.2 1 0.3 23 32.3 12 8.5 5 6.2 5 4.2 4 2.1 3 4.2 a Codesofthe InternationalClassificationof Causesof Death (9threvision)in parentheses. Random effectsestimate; otherwisefixed effects estimates. 0.8 0.9 0.9 1.1 0.7 0.9 1.o 1.4 1.1 0.9 0.7 1.o 0.8 0.9 1.o 0.3 1.2 1.a 0.8 1.4 0.8 1.1 2.4 1.4 0.7-0.0 0.8-0.9 0.61.2 0.6-2.lb 0.3-1.5 0.8-1 .I 0.7-1.5 0.4-2.0 0.8-1.5 0.4-2.2 0.5-1.1 0.6-1.6 0.8-0.9 0.7-1 .O 0.8-1.2 0.0-2.1 0.7-2.0 0.2-20.8 05-12 0.a-2.6 0.3-2.2 0.4-3.1 0.74.0 0.4-5.7 t0.00001 0.16 0.48 0.00062 0.58 0.33 0.47 0.18 0.20 0.23 0.53 0.12 0.11 0.21 0.97 0.81 0.61 0.26 0.28 0.26 0.49 0.62 0.11 0.17 74 Scand J Work Environ Health 1998,vol24, suppl2 . cancer sites are of special interest because of the results studies, as evidenced by the wide confidence intervals in of previous human or animal studies or, for bladder the early studies. The recent studies of Blair et a1 ( 3 3 , cancer, because of our overall meta-analysis findings. Wood et a1 (36), Swaen et a1 (37), and Benn & Osborne Total mortality (38) all had narrow confidenceintervals and the SMR values were close LO 1.O. The mRR for all the studies was 0.9 Figure 1 presents the SMR values for total mortality for (95%.CI 0.9-1.1). each study with their confidence intervals. The studies We examined the cumulative relative risk by the date are arranged by the year of publication or, if unpublished, of the study, as shown in figure 3. Before 1992, the cu- by the year of completion.With the exception of the study mulative relative risk for lung cancer mortality among of Zhou & Wang (31), all the studies had SMR values acrylonitrile workers was slightly greater than 1.O. With equal to or less than 1.O. Zhou & Wang (31) did not de- the completion of the 4 large studies in 1997, the confi- scribe the methods of follow-up used in their study and dence interval was narrow, and the cumulative SMR was stated that the death information may not be comparable below 1.0. to the national population. However, the remaining stud- Latency is the term often applied to the period between ies, though somewhat similar in results on an absolute initial exposure and death from a disease. Most occupa- basis, still showed considerable heterogeneity tional carcinogens do not show increased risk for 15 or (P=0.00004).The US studies were similar (P=0.43) with 20 years after first exposure (44). Eight studies consid- an mRR of 0.7 (95%CI 0.7-0.7). The non-US studies, ered latency periods of 15 years or longer. The studies however, still exhibited some variability (P=0.02),with which considered latency had an mRR of 1.0 (95%CI an mRR of 0.8 (95%CI 0.8-0.9). 0.9-1.1)compared with an mRR of 0.9 (95%CI 0.7- Lung cancer 1.1)for the studieswhich did not. Only the studiesof Blair et al(35) (RR1.3,95%CI 1.0-1.63)andDelzell & Mon- The SMR values for lung cancer by study are shown in son (22) (SMR 1.7,95%CI 0.7-3.5)had elevated rates figure 2. The early studies were smaller than the 4 recent in the longest latency category. The 6 other studies Kelssoibach Theiu Delull o# Herman - Keiuelbach I - Th0i.r c Herman I ott, Dsbll I I - I - I 11.1 I Burke (Memphis) Burke (Beaumont) Mutnngelo Zhou Eenn Swam Wwd Biair Burke(Memphis) , - Burke (Beaumont) ,Mutnngelo - I Benn Smen Uc-.l I-- Wood m Blair IC* I I Y.tr-RR 0 1 Relative Risk Y.tr-RR 0 mi I I 12 I I 3 Relative Risk I 4 I Figure 1. Relative risk for all causes of death (studies listed by first Figure 2. Relative risk for lung cancer death (studies listed by first 1 author). author). Scand J WorkEnvironHealth 7998,~0124s,uppl2 75 Studies of acrylonitrileworkers - -2 , 1.8 -- 1.6 -- 1.4 -- - r I - - - - - - -1.2 -I 0.8-- .*I * .I, 11 I' * * 79- 1 1 - I 1I 11-I1- k 0.6 -- I 0.4 -- 0.2 -- o.:::::::I:I:::::::::I mhdia fggggg?iffgggggfgfg Figure 3. Cumulative relative risk and 95% confidence intervals for lung cancer by year of study. I - ,Keisoeibnh ,,nul88 - D.lzdi I, ,Manh ,,Burke (Memphis) Burke (Beaumont) c . '"I$,Maatmn~8lo Wood 11.3 I I I 15.5 50.4 52.7 I II I I I I I I I 1 , I l l 1 I l l 1I t 0 1 2 3 4 5 6 7 8 910 Relative Risk Figure 4. Relative risk for prostate cancer death (studies listed by first author). reported SMR values I1.O for this category. The mRR for this category was near null (mRR 1.2,95%CI 1.0-1.4, heterogeneity P-valued.52). Several of the acrylonitrilestudies examined worker mortality rates by level of exposure. These analyses are important to separate workers with low or brief exposures from more highly exposed workers who are obviously more relevant for causal inference. Unfortunately,only 7 studies examined cancer risk by level of exposure, and most of these evaluations were for lung cancer. These 7 studies (18, 22, 33, 35-38), which present data for 76 Scand J Work Environ Health 1998,vol24, suppl2 highly exposed workers, had null rates for lung cancer (mRR 1.0,95% CI 0.9-1.1), in comparison with a deficit for lung cancer (mRR 0.7, 95% CI 0.4-1.4) in the studies that did not examine workers with higher exposure. The highest-exposed workers in the 7 studies produced an mRR of 1.2(95% CI 1.0-1.5, heterogeneity Pvalue=O.O6). None of the studies found a trend with exposure level. The aggregation of the results of the highest-ranked exposure group in studies should be viewed as a crude overview, since the respective categories in various studies may represent different exposurelevels and the pooled results could potentially be confounded. Three studies (35-38) estimated exposure levels in parts per million (ppm), which allowed us to examine workers with comparable high exposures. The 3 studies which estimated the parts-per-million levels of exposure had an mRR of 0.9 (95% CI 0.8-1.0) for lung cancer in comparisonwith an mRR of 1.1 (95% CI 0.9-1.4) for the studies which did not estimate parts-per-millionlevels of exposure.We combined the 8ppm-years category in the Blair et al study ( 3 3 , the 10-50,50-100, and 100ppm-year category in the Wood et a1 study (36), and the 10ppm-year category in the Swaen et a1 study (37). The mRR for these studies was 1.1 (95% CI 0.9-1.4, heterogeneity P-value=O.12). Prostate cancer The relative risks for prostate cancer mortality for individual studies is shown in figure 4. The studies of Theiss et a1 (18), Delzell & Monson (22), and Burke (Memphis plant) reported no prostate cancers, but they had wide confidence intervals. The studies of Keisselbach et a1 (15) and Wood et a1 (36) reported small excesses, and Burke (Beaumont works) reported an excess based on a single case. The large studies of Blair et a1 (35) and Swaen et a1 (37) reported slight deficits of prostate cancer. The mRR for the prostate cancer mortality was 1.0 (95% CI 0.71.5). Only the Blair et al(35), Swaen et a1 (37), and Wood et al(36) stuhes reported exposurelevel analyses for prostate cancer risks, perhaps because most other studies had no more than 2 prostate cancer deaths. The mRR was 1.O (95% CI 0.5-1.8, heterogeneity P-value=0.60)for the cumulative exposure level of 8 ppm in the Blair et al study (35) and 10 ppm in the Swaen et a1 (37) and Wood et a1 (36) studies. None of these studies showed an increasing risk with increasing exposure.The single nested case-referent study by Marsh (23) reported duration of exposure for prostate cancer cases and referents. There were no cases or referents in the highest exposure category of 10 years of exposure. The 2 unpublished studies by Burke (Memphis plant and Beaumont works), which reported prostate cancer mortality, reported 1 death from prostate cancer (mRR 3.9) in comparisonwith an mRR of 1.0for the published studies (data not shown).However, we also found some evidence of failing to report relative risks of less than l .O in the published studies. Two authors shared with us the prostate cancer results that were omitted from their original reports (18,33).Taken together, there were 2 observed deaths and 2.1 expected. The Wood et al study (36) found 12prostate cancer cases versus 7.6 observed. The other 2 studies which examined incidence found no cancer cases, with 0.8 and 0.1 expected cases, respectively.The Wood et al(36) updcte of the earlier DuPont studies (24,26) found only 1 new incident case versus 3.9 expected [standardized incidence ratio (SIR) 0.3, 95% CI 0.0-1.41 in the update period. The Chen et al(26) and O'Berg et al(24) studies reported that most cases of prostate cancer occurred in 1975 to 1983.Therefore, the cases of prostate cancer were limited in time. In addition, no trend with exposure level was observed in these studies. Brain cancer The studies of Keisselbach et a1 (15), Herman (unpublished), Burke (unpublished, Memphis plant), Mastrangelo (33), Swaen et a1 (37), and Wood et a1 (36) reported relative risks for brain and central nervous system cancers in excess of 1.0. (See figure 5.) Theiss et al (18), Delzell & Monson (22), Burke (unpublished,Beaumont works), Thomas et a1 (25), and Blair et a1 (35) reported relative risks of 1.O. The mRR for brain cancer was 1.2 (95%CI 0.8-1.7).Only 3 studies reported brain cancer rates by exposure level (25, 35, 37). There was no increase in risk with level of exposure in any of these studies. Although the estimates were imprecise, the relative risks were higher in the unpublished investigations (mRR 2.3,95%CI 0.6-9.3) than in the published studies (mRR 1.1,95%CI 0.7-13.However, as with prostate cancer, there was also a tendency to not report SMR values of e1.O. Studies reporting expected deaths had an mRR of 2.2 (95%CI 0.7-6.4) in comparison with an mRR of 1.0 (95% CI 0.7-1.4) for studies not reporting expected deaths. Bladder cancer There were 10 studies which evaluated bladder cancer (figure 6). The early studies of Kiesselbach et a1(15), Theiss et al(18), and Delzell & Monson (22) all reported S M R values of >1.0. The 2 studies by Burke (unpublished, Memphis plant and Beaumont works) and the study by Mastrangelo (33) found no cases although few were expected. The case-referent study of Siemiatycki et a1 (34) reported a small excess, and the 3 largest studies of Swaen et a1 (37), Blair et a1 ( 3 3 , and Wood et a1 (36) reported SMR values close to 1.0.Blair et a1(35), the only authors to report exposure levels for this cancer, found no increased risk with increasing exposure level. As with brain cancer, the studies reporting expected deaths had a larger mRR (2.9,95%CI 0.9-9.7) than those which failed to report expected values (mRR 1.3,95%CI 0.8-1.9). Some of the most potent occupational causes of bladder cancer are aromatic amines (44).Three studies reported the presence of aromatic amines in the plant environment (15,18,22).The mRR for bladder cancer for studies reporting potential exposureto aromatic amines was - ,.K e h l t u c h Theiu II ,Herman D.b.ll I, -- I -- I 12.7 10.9 Burke (Memphis) I 16.7 Burke (Beaumont) II T 4Thomas c ,Mastrangelo - 23.8 14.6 swam I , -wood 1 --,Blair I Meta-RR ,cI IIIIIIIII IIIIIIII 0 1 2 3 4 5 6 7 8 9 1Q Relative Risk Figure 5. Relative risk for brain cancer death (studies listed by first author). Scand J Work Environ Health 1998,vol24, suppl2 77 Studiesof acrylonitrileworkers Kelawlbach I Theiu - D.lull Burke (Memphis) Burke (Beaumont) Muttangolo Skmiatfcki 17.4 11.8 14.4 15.4 32.5 17.1 Smen Wood Blair Metr-RR 0 I 2 3 4 5 6 7 8 9 10 Relative Risk Figure 6. Relative risk for bladder cancer death (studies listed by first author). 4.5 (95% CI 1.8-10.9) based on 7 deaths. The mRR for studies with no aromatic amines reported was 0.9 (95% CI 0.5-1.5). These data are consistentwith the possibility of confounding, but only 1 of the 3 studies indicated that the bladder cancer decedents (3 deaths) had exposure to aromatic amines (15). Other potential confounding exposures Since butadiene exposure occurred in some of the plants in various studies, we stratified for this potential exposure. We found no difference in the cancer rates from plants which had potential butadiene exposure from the plants which did not. Discussion Most of the available epidemiologic studies of acrylonitrile workers are cohort mortality investigations.The cohort studies in ths review range in size from 100to 25 460 workers. Taken together, the studies include 41 135 exposed workers and 2769 decedents. The levels of exposure in many of the studies are uncertain since monitoring for acrylonitrile was not 78 Scand J Work Environ Health 1998,voI24, suppl2 available before the mid-1970s. This lack of information made it difficult to focus our analysis on the highest-ex- posed workers, who would be the most relevant for an evaluation of acrylonitrile carcinogenicity. We examined the impact of publication bias on the results.The published studies were, on the average,4 times larger and followed workers for 7 years longer. However, there was little difference between the published and unpublished studies in completeness of vital status followup (97% versus 96%), completeness of death ascertainment (98% versus 96%), or mean duration of exposure (6.9 years to 7.5 years). The published studies had lower total mortality than the unpublished studies. The mRR for all causes was 0.6 (95% CI 0 . 5 4 . 7 ) for the unpublished studies and 0.8 (95% CI 0 . 7 4 . 9 ) for the published studies. The total cancer rates were slightly higher for the published studies (mRR 0.9, 95% CI 0.8-0.9) than for the unpublished studies (mRR 0.8,95% CI 0.4-1.2). While we found some evidence of publication bias in the literature, it did not greatly affect the risk estimates for individual cancers. There was also an apparent tendency not to report null findings for certain cancer sites. For instance, while 10of the cohortmortality studies reported lung cancer rates, only 4 of these studies reported prostate cancer rates. This finding lends some credence to the view that null findings may be omitted frompapers, especially when there is no reason to believe that an exposuredisease relationship is plausible, and that this omission exerts a positive bias in quantitative reviews of the literature. However, the omission of these null findings did not have a great impact on the risk estimates. We also found that the early studies tended to find relative risks of >1.0 for lung cancer. This could be a chance finding or it could reflect an early preference for the publication of positive findings. Taken together, these factors indicate that there was a slight positive bias in the published studies, but this bias did not have a great impact on the risk estimates for individual cancers. As in a previous review of acrylonitrile studies, we found no excess of all cancer or lung cancer among the acrylonitrile workers (2).In addition, our results are similar to those reported by Rothman (2), even when recently available studies are considered and analyses are conducted with respect to exposure level and latency periods. We were unable in our analyses to take the smoking habits of workers into account, which is important for evaluating lung cancer. Blair et al(35) systematicallyevaluated the smoking levels of workers. Their study of 25 460 workers found that the proportion of cigarette smokers was larger among workers exposed to acrylonitrile and that the proportion of smokers increased with increasing cumulative exposure category. If the findings of this large study can be generalized to the other studies, smoking may be an important confounder to be considered. 1 There was some indication of excess bladder cancer in 3 studies, a finding not reported previously. However, the excess seems to be restricted to plants with potential exposure to aromatic amines, and, therefore, it is unlikely t o be related to acrylonitrile exposure. No trend with exposure level was observed (35). The siight excess of prostate cancer incidencereported for 1 population by O'Berg et a1 (24), Chen et al(26), and Wood et a1 (36) has raised concern that exposure to acrylonitrile may increase prostate cancer incidence (6). However, there is no increase in cancer rates with increasing exposure, and this finding has not been seen in mortality studies (15, 18,22,33,35-37) or in the incidence studies (both unpublished studies by Burke, Memphs plant and Beaumont works). The excess of prostate cancer in this 1 study was limited to a narrow reporting period (ie, 1978-1983). A deficit was observed (SIR 0.3, 95% CI 0.0-1.4) for 1984-1991. Taken together, these findings do not seem indicative of a causal association. Bias or chance are a more likely explanation. Acrylonitrile is a multisite carcinogen in rats chronically exposed to 20 ppm. However, findings from humans show null or near null findings when studies are considered together. There were no patterns of risk indicative of a causal associationin our study. Such results suggest, at most, a small increase in risk among workers for the length of follow-up included in the available studies. Acknowledgments We are grateful to Dr Gerard Swaen, Sir Richard Doll, and 3 anonymous reviewers for providing constructive suggestionson an earlier draft of this paper. We also thank Dr Larry Holden and Susan Riordan of Monsanto for as- sisting with the calculationsand Diane Bowens and Janet Delaney for their data entry. References 1. Anonymous. Production soared in most chemical sectors. Chem Eng News 1 9 9 5 ; 7 3 : 3 8 4 . 2. Rothman KJ. Cancer occurrence among workers exposed to acrylonitrile. Scand J Work Environ Health. 1994;20:313- 21. 3. International Agency for Research on Cancer (IARC). Eval- uation of the carcinogenic risk of chemical to humans: some monomers, plastics, synthetic elastomers, and acrolein. IARC, Lyon, 1979. IARC scientific publication, vol 19. 4. World Health Organization (WHO). Acrylonitrile. Geneva: WHO 1983. Environmental health criteria, no 28. 5. American Industrial Hygiene Association. Odor threshold for chemicals with established occupational health stand- ards. Akron (OH): American Industrial Association, 1989. 6. Agency for Toxic Substances and Disease Registry (ATSDR). Toxicological profile for acrylonitrile TP-9002. Atlanta (GA): ATSDR, 1990. 7. Maltoni C, Ciliberti A, DiMaio V. Carcinogenicity bioassays on r t s of acrylonitrile administered by inhalation and ingestion. Med Lav 1977;68:401-11. 8. Koerselman W, van der Graaf M. Acrylonitrile: a suspected human carcinogen. Int Arch Occup Environ Health 1984;54:3 17-24. 9. Guirguis SS, Cohen MB, Rajhans GS. A review of health risks in acrylonitrile industry. G Ital Med Lav 1984;6:8793. 10. Strother DE, Mast RW, Kraska RC, Frankos V. Acrylonitrile as a carcinogen: research needs for better risk assessment. Ann NY Acad Sci 1988;534:169-78. 11. United States Environmental Protection Agency. Health assessment document for acrylonitrile: general review and summary. Sci Total Environ 1990;90337-52. 12. Doll R. Hazards of cancer in the chemical industry. BIBRA Bull 1991;30:183-88. 13. Ward CE, Starr TB. Comparison of cancer risks projected from animal bioassays to epidemiologic studies of acrylonitrile-exposed workers. Regul Toxic01 Pharmacol 1993;18:214-32. 14. Blair A, Kazerouni N. Reactive chemicals and cancer. Cancer Causes Control 1997;8:476-93. 15. Kiesselbach N, Korallus U, Lange HJ,Ness A, Zwingers T. Bayer acrylonitrile study 1977: central journal for industrial medicine, protection of labor, industrial hygiene, and ergonomics. Heidelberg: Dr C Haefner Verlag GmbH, 1980. 16. O'Berg MT. Epidemiologic study of cancer incidence and mortality among workers exposed to acrylonitrile at a chemical plant which produces textile fibers. Ann Arbor (MI): University Microfilms International, 1980. 17. O'Berg MT. Epidemiologic study of workers exposed to acrylonitrile. J Occup Med 1980;22:245-52. 18. Theiss AM, Frentzel-Beyme R, Link R, Wild H. Mortalitatsstudie bei chemiefacharbeitern verschiedener produktionsbetriebe mit exposition auch gegenuber acrylnitril. Zentralbl Arbeitsmed Arbeitssch Prophyl Ergonomie 1980;30:259-67. 19. Ott MG, Kolesar RC, Scharnweber HC, et al. A mortality survey of employees engaged in the development or manufacture of styrene-based products. J Occup Med 1980;22:445-60. 20. Werner JB,Carter JT. Mortality of United Kingdom acrylonitrile polymerization workers. Br J Ind Med 1981;38:24753. 21. Waxweiller RJ, Smith AH, Falk H, Tyroler HA. Excess lung cancer risk in a synthetic chemicals plant. Environ Health Persect 1981;41:159-64. 22. Delzell E, Monson RR. Mortality among rubber workers, VI: men with potential exposure to acrylonitrile. J Occup Med 1982;24:767-9. 23. Marsh GM. Mortality among workers from a plastics producing plant: a matched case-control study nested in a retrospective cohort study. J Occup Med 1983;25:219-30. 24. O'Berg MT, Chen JL, Burke CA, et al. Epidemiologic study of workers exposed to acrylonitrile: an update. J Occup Med 1985;27:835-40. 25. Thomas TL, Stewart PA, Stermhagen A, Correa P, Norman SA, Bleecker ML, Hoover RN. Risk of astrocytic brain tumors associated with occupational exposures. Scand J ScandJ Work Environ Health 1998,vol24, suppl2 79 Studies of acrylonitrileworkers Work Environ Health 1987;13:417-23. TF, et al. Mortality of industrial workers exposed to acrylo- 26. Chen JL, Walrath J, O'Berg MT, Burke CA, Pell S. Cancer nitrile. Scand J Work Environ Health 1998:24 suppl 2:25- incidence and mortality among workers exposed to acrylo- 41. nitrile. J Occup Med 1987;11:157-63. 36. Wood SM, Buffler PA, Burau K. Krivanek N. Mortality and 27. Chen JL, Fayerweather WE, Pell S. Cancer incidence of morbidity of workers exposed to acrylonitrile in fiber pro- workers exposed to dimethylformamide and/or acryloni- duction. Scand J Work Environ Health 1998;24suppl2:54- trile. J Occup Med 1988;30813-8. 62. 28. Chen JL, Fayerweather WE, Pell S. Mortality study of 37. Swaen GMH, Bloemen LJN, Twisk J, Scheffers T, Slangen workers exposed to dimethylformamide and/or acryloni- JJM, Collins JJ, et al. Mortality update of workers exposed trile. J Occup Med 1988;30:819-22. to acrylonitrile in The Netherlands. Scand J Work Environ 29. Ott MG, Teta MJ, Greenberg HL. Lymphatic and hemat- Health 1998;24 suppl 2:10-6. opoietic tissue cancer in a chemical manufacturing environment. Am J Ind Med 1989;16:631--43. 38. Benn T, Osborne K. Mortality of United Kingdom acryloni- -trile workers an extended and updated study. Scand J 30. Collins JJ, Page LC, Caporossi JC, Utidjian HMD. Mortali- Work Environ Health 1998;24 suppl 2: 17-24. ty patterns among employees exposed to acrylonitrile. J 39. Greenland S. Quantitative methods in the review of epide- OCCUPMed 1989;31:368-71. miologic literature. Epidemiol Rev 1987;9:1-30. 3 1. Zhou B, Wang T. Historical cohort study of causes of death 40. Dickersin K, Berlin JA. Meta-analysis: state of the science. in a chemical fiber factory. J Chin Med Univ 1991;20:35- Epidemiol Rev 1992;14:154-76. I . 41. SAS Institute Inc. SAS user's guide: basics, version 5 32. Swaen GMH, Bloemen LJN, Slangen JJM, Twisk J, Schef- edition. Cary, NC: SAS Institute Inc, 1993. fers T, Sturmans F. Mortality of workers exposed to acrylo- 42. Shadish WR, Haddock CK. In: Cooper H, Hedges C, edi- nitrile. J Occup Med 199234801-9. tors, The handbook of research synthesis. New York (NY): 33. Mastrangelo G, Serena R, Maszia V. Mortality from tu- Russell Sage Foundation, 1994:261-81. yun:t mours in workers in an acrylic fibre factory. Occup Med 43. Breslow NE, Day NE. Statistical methods in cancer re- ,, II 5 r.' 1993;43:1568. 34. Siemiatycki J, Dewar R, Nadon L, Gerin M. Occupational sear& vol 11 (The design and analysis of cohort studies). Lyon, International Agency for Research on Cancer, i: risk factors for bladder cancer: results from a case-control 1987:113-5. study in Montreal, Quebec, Canada. Am J Epidemiol 44. Cole P, Goldman MB. Occupation. In: Fraumeni JF,editor, 1994;140:1061-80. Persons at high risk of cancer: an approach to cancer etiolo- 35. Blair A, Stewart PA, Zaebst DD, Pottern L, Zey JN, Bloom gy and control. New York (NY):Academic Press, 1976. 80 Scand J Work Environ Health 1998,vol24, suppl2