Risk of astrocytic brain tumors associated with occupational chemical exposures. A case-referent study.

HOOVER RN. Risk of astrocytic brain tumors associated with occupational chemical exposures: A case referent study. Scand J Work Environ Health 13 (1987) 417-423. A case-referent study was conducted on the risk of brain tumors among workers exposed to organic chemicals in petroleum refining and chemical manufacturing. Brain tumor cases in northern New Jersey, Philadelphia, and the Gulf Coast of Louisiana were identified from death certificates of a recent three-year period. The cases (N = 300) were white men aged ~ 30 years with a confirmed diagnosis of glioblastoma multiforme, astrocytoma, or a mixed glioma with astrocytic cells. The referents (N = 386) were white men who died from causes other than brain tumor, epilepsy, cerebrovascular disease, suicide, or homicide and were frequency-matched with the cases on age at death, year of death, and study area. Next-of-kin were interviewed for complete occupational histories. No statistically significantly elevated odds ratios (OR) were associated with employment in the chemical industry. The risk of astrocytic tumors was elevated among the subjects with production or maintenance jobs in petroleum refining (OR 1.7, 95 % confidence interval 0.7 - 4.2); however, it decreased with duration employed. There were nonsignificant excessrisks of astrocytic tumors among the men exposed to cutting fluids (OR 1.6) or organic solvents (OR 1.3), and also among the subjects exposed to lubricating oils (OR 1.4), organic solvents (OR 1.5), or cutting fluids (OR 1.8) for ~ 20 years.

though expected numbers could not be calculated for specific ceIl types, many of the earlier studies had suggested that the elevated risks were cell-type specific and were limited to glioblastoma multi forme and astrocytoma (25).
A case-referent study was conducted in northern New Jersey, Philadelphia (Pennsylvania), and the Gulf Coast of Louisiana to investigate findings of hypothesis-generating studies and to evaluate other risk factors for brain tumors. The present analyses focused on risks for astrocytic tumors (glioblastoma multiforme, astrocytoma, and mixed glioma with astrocytic cells) among men employed in the petroleum refining and chemical manufacturing industries and evaluated risks for particular organic chemical exposures thought to be associated with excess brain tumor risk.

Subjects and methods
Because this investigation was planned to test hypotheses regarding the association between brain tumor mortality risk and employment in the petroleum and chemical industries, study areas were chosen on the basis of employment patterns by industry. Three geographic areas with consistently high proportions of the workforce employed in the petroleum refining and chemical manufacturing industries combined were northern New Jersey, the Philadelphia, Pennsylvania, standard metropolitan statistical area, and the Gulf Coast area of Louisiana. Each study area included at least one standard metropolitan statistical area and surrounding counties in which a substantial proportion (5 to 190/0) of the workforce was likely to have been employed in the industries of interest.
Permission was obtained from state vital records offices to review causes of death among usual residents of the selected study areas and to utilize information from death certificates of cases and referents to contact their next-of-kin . Cases were white men, aged 30 years or older, who died of brain or other central nervous system tumors [International Classification of Disease 9th Revision (lCD-9) (29) codes 191, 192,225, 239.7] during a three-year period and who were usual residents of counties in the selected study areas. The study period for Louisiana was 1 January 1978 through 30 June 1980, and for the other two study areas it was I January 1979 through 31 December 1981. The total number of cases ascertained from computer listings was 741.
With the use of a matching ratio of 1 : 1, referents were randomly selected from white male residents who had died of causes other than brain tumor, cerebrovascular disease (lCD-9 codes 430-438) , epilepsy (lCD-9 code 345), suicide, and homicide. Cerebrovascular disease and epilepsy were excluded to decrease the likelihood of including undiagnosed brain tumor cases in the reference series. Suicide and homicide were excluded due to the sen sitive nature of cause of death and the potential difficulties in obtaining responses from next-of-kin. The referents were frequencymatched to the cases by age and year of death. The resultant age and year of death distributions for the cases and referents were nearly identical.
Death certificates were not obtained for one casc and four referents who died in states outside the study areas; therefore a total of I 478 subjects were left whose next-of-kin could be contacted for interview. Informants listed on the death certificates were contacted first by letter and next by telephone to ascertain the identity and address of the next-of-kin. The next-of-kin was chosen according to the following hierarchy: (i) spouse (65 %), (ii) adult child or parent (22 %) , (iii) sibling (8 %), (iv) other (5 %). Next-ofkin of 212 study subjects (86 cases, 126 referents) either could not be traced or could not be identified from information available on the death certificates. About 70 % [N = 869, 483 cases (74 %), 386 referents (63 %)] of the I 266 next-of-kin who were located agreed to be interviewed. The final study group was a sufficient size to detect an excess brain tumor risk of 2.0 or greater among men employed in the suspect "high risk" industries with a statistical power of 70 %.
In order to assess potential response bias, wc compared the demographic characteristics on the death certificates of subjects with completed interviews (com-418 pletes) with those whose next-of-kin could not be located (untraced) and those whose next-of-kin refused to be interviewed (refusals). There were slightly lower proportions of white-collar workers and married men among the untraced than among the refusals and completes. Because their next-of-kin were difficult to locate, the untraced had a slightly higher proportion of subjects who died in the earliest years of the study period. There was little variation between refusals and completes with respect to age , marital status, and job status (ie, blue collar versus white collar). Since there were also no differences in the proportions of cases and referents by these variables, it could be assumed that there was little or no response bias affecting these factors . The response rate varied substantially by study area, being highest in Louisiana (81 %) , and might have been affected more by regional social attitudes than by other variables.
Because earlier studies suggested that an excess risk of brain tumors in the petroleum and chemical industries was limited to specific cell types (25), the present analyses included study subjects as cases only if the best diagnostic information available indicated the presence of an astrocytic brain tumor (astrocytoma, glioblastoma multiforme, mixed glioma with astrocytic cells). Hospital records were sought for the 483 cases with completed interviews. Forty-eight of the cases did not have primary brain tumors and were excluded from further consideration. The remaining cases consisted of 300 men with astrocytic tumors and 135 with other (N = 90) or undetermined (N = 45) cell types. Two hundred and twenty-nine (76 %) of the astrocytic tumors were pathologically confirmed , and the remaining 71 were diagnosed by physicians on the basis of appearance in computerized tomography. The final data set considered in this report consisted of the 300 cases with astrocytic tumors and 386 referents.
The lifetime occupational history data collected in the interview included the following information on each job held by the study subject since the age of 15 years: job title, brief job description, kind of company or organization, company name, job location (city, state), employment dates, and hours worked per week . Information on ethnic background, education, and other possible risk factors for brain tumors, including history of head injuries, seizures, drug usage, famil y history of cancer, and other environmental factors was also collected. Interviewers were blinded regarding the case-referent status of the subject until the end of the interview .
Each job entry in a study subject's work history was assigned a Standard Industrial Classification (SIC) code for industry (16) and a census code for occupation (3). The " high risk" industries considered in these analyses were petroleum refining (SIC 291 -299) and the manufacture of chemicals and allied products (SIC 281-289).
An industrial hygienist (PAS) classified each job according to whether it had potential exposure to organic solvents, lubricating oils, polycyclic aromatic hydrocarbons, formaldehyde, acrylonitrile, vinyl chloride , or phenolic compounds. All coding was done blindly, ie, without knowledge of whether the subject was a case or referent. Jobs considered to have potential exposure to acrylonitrile in th ese data included those in the production of plastics or rubber, but the y were primarily in farming where acrylonitrile may be used as a pesticide. Jobs with exposure to vinyl chloride were production jobs in the plastics indu str y. Formaldehyde-exposed jobs were those that invol ved oil drilling, cloth dying, carpentry, certain laboratory work (eg, biological science and histology laboratorie s), and the manufacture and sale of furniture and clothing. Mechanics, precision metal workers, refinery workers, and truck and bus drivers were assigned lubricating oil exposure, and cutting fluid exposure was assigned primarily to jobs that involved precision metal work. The category of lubricating oils included all oils and hydraulic fluids used to lubricate motors, engines, and equipment, while cutting fluids, a specialized subset of lubricating oils, included only those used to lubricate and cool cutting surfaces in precision metal work . Jobs in laboratories, carpentry, painting , pipefitting, cloth dying , construction, and printing were considered to have potential exposure to phenolic compounds (ie, phenol, cresol, creosote, hydroquinone, phenolic resins, quinone, and resorcinol) . Pol ycyclic aromatic hydrocarbon exposure was assigned to mechani cs, gas station attendants, plumbers, roofers, oil workers, preci sion metal workers, foundr y and steel workers , boiler room operators, and truck and bus dr ivers.
Nearl y all of the jobs that were con sidered expo sed to the aforementioned substances also might have involved contact with organic solvents, and there was co nsiderable overlap between the va rious exposure categories. Because so many subjects were considered exposed to solvents, the industrial hygienist classified expo sure by presumed frequency of solvent use. A job was assumed to have "low" expo sure if it was likely to involve the use of solvents less than once per workday . "Moderate" exposure was assigned to a job if solvent use was likely to have been one to several times per day, while jobs requiring more con stant solvent contact were co nsidered to have "high" exposure.
The unexposed subjects for the analyses by industry included tho se never employed in the petroleum refining or chemical manufacturing industries. In the anal yses by organic chemical substance, a study subject was considered to have been "exposed" to a substance if he ever had a job that might have involved contact with the substance, the unexposed being men who never had a job involving the particular organic chemical substance being considered. Excluding those ever exposed to any of the other substances would have resulted in a reference group that was largely white collar and that might differ from the exposed group in many characteristics. Duration of exposure was estimated from the calculation of the total number of years spent in "exposed " jobs. If dates worked were unknown for a particular job, the job was excluded from the analyses by duration employed.
To estimate the relative risk of brain tumors, we calculated the maximum likelihood estimates of the odds ratio (OR) for brain tumors and the 95 % confidence intervals (95 % CI) (8) for each industry and organic chemical substance of interest. Since the cases and referents were frequency-matched on study area, age, and year of death only to obtain a reference series that was similar to the cases with respect to these factors, an unmatched analysis was conducted and was stratified by potential confounders. Variables assessed for possible confounding were age at death, history of cigarette smoking, history of alcoholism , ethnic background , and highest level of education. Level of education was the only variable which appeared to be related both to brain cancer and the exposures of interest; thus , all analyses were stratified by it. The analyses by organic chemical substance were also stratified by a variable which indicated ever or never exposure to the other chemicals. An odds ratio was considered statistically significant if the 95 % confidence interval did not include 1.0. A chi-square test for linear trend was used to determine whether trends by duration employed were significant (19). Table 1 shows the odds ratios for astrocytic brain tumors among the st udy subjects who were ever em-

419
------------------a Max imum likelihood estimate of the odds ratio stratified by educational class. b The unexposed subjects in each of these analyses were men who had never worked in any of the "hig h risk " industries.   Table 3 shows results for the selected organic chemical substa nces. The OR estimates were slightly elevated for organic solvents (OR 1.3) and cutting fluids (OR 1.6), but neither was statistically significant. The odds ratio for vinyl chloride was 1.3 [omitted from table due to small numbers (3 cases, 6 referents, 95 0J0 CI 0.8 -2.0)]. No excess risk of astrocytic tumor mortality was seen among the men who had jobs that might have involved exposure to acrylonitrile, formaldeh yde, lubricating oils, polycyclic aromatic hydrocarbons, or phenolic compounds.
Th e odd s ratio s for astrocytic tumors by duration of " exposed" jobs are shown in table 4 by organic chemical substance. Astrocytic tumor mo rtalit y risk was the highest in the 20 or more years category for lubricating oils (OR 1.4), organic solvent s (OR 1.5), and cutting fluids (OR 1.8), but for none of these exposures was there an y con sistent or significant tren d by duration. Table 5 shows results for total duration of solvent exposure by the highest level of solvents to which subject s were classified as exposed . The odds ratio was elevated among the subjects whose highest expo sure was low (OR 1.4) or moderate (OR 1.6) and who were expo sed to organic sol vents for 20 or more years. No excess risk was seen among the men classified as having high solvent exposure . However, the number of workers expo sed in each duration category was small.
The findin gs for cutting fluid s were examined further with calculatio ns of OR estimates for specific precision metal work jobs which would involve the exposure (table 6). Overall, the odd s ratio for precision metal work was 1.6 (95 010 ClI.O -2.6). The odd s ratios increased with duration employed for all the precision metal workers combined . The highest risk was seen for tool and die makers . Howe ver , none of the men had this job for 20 or more years. Machinists had an odds ratio of 1.8 after 20 years of emp loyment, but the number of men classified in individual duration categories was sma ll. ployed in petroleum refining or chemical manufacturing . The risk was slightly elevated among all the men ever employed in either of the " high risk " industries, but neither result was statistically significant. Men employed in production and maintenance jo bs in petroleum refining had elevated astro cytic tumor risk, and those only employed in white-collar jobs did not. However, the confidence intervals for the two OR estimates overlapped. In chemical manufacturing, the odds ratios for those employed in production and maintenance jobs were the same as tho se for men employed in white-collar jobs. Analyses of astrocytic tumor risk in petroleum refining by duration employed indicated that risk was highest among those employed in the industry for less than five years (table 2), and the odds ratios decreased with increasing duration employed. This finding was true for all workers combined and for those empl oyed in production an d ma int ena nce . No te: The unexposed subje ct s for each calc ulation were tho se whose jobs had never inv olved exposure to the substance o f interest.

Discussion
Analyses for two "high-risk" industries suggested that there might be an elevated risk of astrocytic brain tum o r mortality associated with ever having a job that involved chemical expo sure in oil refining, but not with chemical manufacturing. These result s wou ld be consistent with report s suggesting elevated brain tumor risk among production and maintenan ce workers in oil refining (21,23,26) and contrary to oth er studies of refinery workers which failed to find excess risk of brain tumors (6,10,14,20,27). Ho wever, none o f the odds ratios in the current stud y were sta tistically significant , and the decreasing risk by dur ation emp loyed casts doubt on the causality of the associa tion. The highest risk occurred among the men who wo rked less than five years (OR 3.9), were hired before the age of Table 4. Odd s rali os for astrocytic tumors amon g subjec ts w hose jo bs ever might have involved exposure to selec ted organic chemi cal su bsta nces by durat io n of " exposed" jo bs . (C = number of case s, R = number of referents) Type of exposure  , Maximum likelihood estimate of the odds ratio stratified by educational class and other exposures. n The unexposed subjects in each of these analyses were those whose jobs never involved exposure to the substance of interest.

t.
The specific durations of exposure exclude the subjects whose duration was unknown. The unexposed subjects in each of these ana lyses we re those whose j obs had never invo lved expos ure to organic so lvents. C The spec if ic du rations of exposure exc lude t he subjects w hose durat ion was unk now n , but these su bj ec t s are inc luded in the " Total." Table 6. Odd s rat ios for ast rocy tic tumors among precisi on metal workers.  C The spe cific du rati on s exposed exclu de the subject s w ho se du rat ion was unknown , bu t thes e subjects are included in the " Tot al." 40 years (0 R 3.9) , and died more than 10 year s after they first worked in the indu stry (OR 3.9). Earlier cluster s of brain tumors seen in particular refinery populat ion s could represent an excess risk due to the presence of a particular exposure or co uld repr esent geo-3 graphic variation in disease. Th ose findin gs were based on hypothesis-generating methodologies (24,26) or very small numbers of deaths (21,23). The onl y large cohort study which sug gested a brain tumor excess among refinery workers had a short follow-up period (21). Other cohort studies did not find brain tumor excesses, but included white-collar workers (6,14,20) or had shor t follow -up period s (10). As noted, our ana lytic st udy also generated little evidence to suggest th at excess brain tumor risk is associated with longterm employment in the petroleum refining industry. Thu s, in the aggregate , as stronger epidemiologic methodologies have been used to study brain tumor risk among refinery workers, the association has weakened. Intensive st udies of refiner y populations are unlikel y to yield an y further insight regarding the etiology of brain tumors if the number of workers exposed to a put at ive causative agent is small. Contrary to the result s of some earlier studies (I, 18), no excess astrocytic tum or mortality risk was associated in our stud y with employment in the chemical manufacturing industry. However, in the present study, the numbers of study subjects ever employed in specific types of chemical plants were too small to evaluate subsets of this complex industry.
There was an elevated risk of mortality from astroeyrie tumors that was almo st twofold among subjects who worked in jobs with probable exposure to cutt ing fluids for more than 20 years . Precision metal workers also had elevated astr ocytic tumor risk . Although not sta tist ically significa nt, these findings would be consistent with those of previous studies that suggested an elevated bra in tum or risk amo ng machinist s (4,5,9,22,28).
Astrocytic tumor mortality risk among men exposed to organi c solvents was highest among those exposed for 20 or more years and whose highest expos ure was classified as low or moderate. Although there was no excess risk fo r men who were classified as ever having high solvent exposure, there were relati vely few subjects in the hea vy use catego ry. Th ese findings could be co nsistent with numerou s other epidem iologic studies that have detected elevated risk of brain tumors among workers exposed to organic solvents (25). Many organic solvents are highly lipid-soluble, pass readily through the blood-brain barrie r, are known to be neurot oxic (7), and could be hum an car cinogens (7, I I).
A number of potential methodologic concerns is raised by a study based on data from death certificates. First is the concern about completeness of case ascerta inment. Thi s was not considered a major limitation becau se the tumors of pa rticular interest were the astrocytic cell types, which a re ra pidly fat al (2). It has been estimated th at in th e United States 97 % of the incident cases of all malignant br ain tumors would eventually be ident ified fro m death cert ificates (17). Thu s it is unli kely that a significa nt proportion of incident cases of astro cytic tum ors was missing from the present series .
A second limitation of this investigation was the use of death cert ifica tes to select the refe rence series. Although the referent s were repre sent ative of the population of white men who died at the ages selected, they 422 might not ha ve been repr esentative of the population in genera l with respect to characteristics th at lead to early death . A comparison of two reference series selected for a study in Minne sota indicated that deceased referents were heavier smokers and heavier consumers of alcohol and dru gs than the living reference series. However, there were no substantial differences between tho se two reference series in usual occupation (12,13).
Finally, the anal yses for organic chemical substances sho uld be interpreted with ca ution because of the crudit y o f the method used to estimate expo sures. The intent of this exposure classification was merely to refine hypotheses regarding specific expo sures that may be implicated in excess brain tumor risk. Expo sures were assigned to jobs based onjob title, industry, and description of duties provided by next-of-kin; therefore, some misclassification is likely. A problem in identifying exposures based on job title is the uncertainty of whether the exposure actually occurred. For some substa nces, the problem is likely to be minimal. Fo r example, cutting fluid expo sure was assigned pr ima rily to job s that involve precision metal work , and it is highly likely th at most of the men who worked in these jo bs would have had some co ntac t with cutting fluids. On the other hand , acrylon itrile exposure was assigned to an y job where exposure was possible bu t not necessarily likely (eg, farm ers who might have exposure to acrylonitrile, which is often used in pesticides). Therefore, some of our classifications included men who might never have been expo sed to the substance. Because coding was don e withou t knowl edge of the cau se of death of the stud y subjects, any rnisclassificat ion is likely to have been random , thu s tending to underestimate any true differences between the cases and referents.
In summa ry, these analyses failed to pro vide persuasive evidence of a causal association between astro cytic tumor mortality risk and employment in the industries considered to be "high risk." This result may have been due to the small number of study subjects employed in these industries, even though the study area s were chosen to maximize the proportion of the genera l population employed in the petroleum and chemical industries. If a particular chemical substa nce present in these indu stries is responsible for excess bra in tumor risk, the number of person s actually exposed may be too small to detect an elevated risk when these ind ustries are studied as a whole, without attention to part icular expo sure s. Altern atively, the previou s observatio ns of cluster s in particular plan ts ma y be du e to cha nce . This study did indi cate th at some exposures common to these indu stries were associat ed with elevat ed risk of astrocytic tumor mortalit y, and future investigations could focu s on solvents , cutt ing oils, and oth er exposures that ma y be commo n to the high-risk groups.