Male bladder cancer risk and occupational exposure according to a job-exposure matrix—a case-control study in British

Objectives The authors investigated the risk of bladder cancer in association with exposure to over 12 000 occupational chemical agents, complex mixtures, and other substances (hereafter referred to as chemical agents). Methods Adult males diagnosed with cancer between 1983 and 1990 in British Columbia, Canada, were surveyed. Detailed occupational histories and confounding information was provided by a self-administered questionnaire. Cancer controls were matched to bladder cancer cases, resulting in 1062 cases and 8057 controls for the analysis. An extensive United-States-based job-exposure matrix was applied to estimate cumulative exposure to occupational chemical agents. Odds ratios for bladder cancer due to exposure to chemical agents were estimated via conditional logistic regression analyses, adjusted for important confounders. Results A significantly (P<0.05) increased risk was detected for ever exposure to 635 chemical agents, and 341 chemical agents exhibited a significantly increasing dose–response relationship. Adjustment for multiple comparisons resulted in a subset of 29 chemical agents that continued to show significant results. A principal components analysis classified these 29 chemical agents into five independent groups, distinguished mainly by job. Exposures to these chemical agents were largely due to employment in the logging and construction industries and occupations involving motor vehicles. Consistent results were observed for bladder carcinogens reported in the literature. chemical agents cancer. chemical agents results and document a risk for specific chemical agents not previously reported.

The second most important risk factor is occupation, which is estimated to be responsible for up to 24% of all bladder cancers (4). Occupational exposure to the following chemical agents is associated with an increased risk of bladder cancer: aromatic amines, paints and solvents, dyes, leather dust, polycyclic aromatic hydrocarbons (PAH), motor exhaust, and chlorinated hydrocarbons (1,(5)(6)(7)(8). Further epidemiologic research is required for many other potential bladder carcinogens.
Excess bladder cancer incidence was previously observed within specific occupations and industries in a case-control study in British Columbia, Canada (9).

Richardson et al
Using the same cancer register-based dataset, we investigated which specific chemical agents may be responsible for the excess risks observed. Cumulative exposures to 12 456 different occupational chemical agents were estimated with the use of a job-exposure matrix from the United States (US) and the participants' job histories. This article summarizes the findings, including results for the substances classified by the International Agency for Cancer Research (IARC) as definite or probable carcinogens for bladder cancer (6,7).

Study population
The study design has been described elsewhere (9,10) and is summarized here. A self-administered questionnaire was mailed to male cancer patients ascertained by the British Columbia Cancer Registry, aged 20 years and older at the time of diagnosis, during the period 1983 to 1990. Although occupational bladder cancer has been suspected for women, only data for men were available for this analysis. The questionnaire requested information on sociodemographic factors such as ethnicity, education, marital status, and lifetime smoking and drinking habits. Detailed lifetime job descriptions were requested for all jobs of at least 1 year in duration, including the location and duration (years) of each job, and whether the job was full-time or part-time.
Questionnaires were sent to 25 726 eligible men, and 15 463 forms were returned (60.1%). Of the respondents, 1129 had bladder cancer. Histological confirmation of the diagnosis was obtained for all of the cases. The anatomic tumor site was coded using the ninth revision of the International Classification of Diseases (11) and grouped into three-digit categories for the analysis. Job descriptions were coded according to the four-digit 1980 Canadian Standard Industrial Classification (CSIC) and the 1980 Standard Occupational Classification (CSOC) codes (12,13).

Exposure assessment
Exposure to various occupational substances was assessed with the use of data from the National Occupational Exposure Survey (NOES) that was based on the job-exposure matrix (JEM) developed by the United States (US) National Institute for Occupational Safety and Health (14,15). The NOES-JEM is essentially a three-dimensional array with US industry codes on one axis, US occupational codes on another axis, and five-digit substance codes on a third axis. The elements of the array indicate the probability of exposure to a specific substance in a specific job. As calendar year is not included in the matrix, the probability of exposure is independent of time. Jobs were classified according to the US 1980 census of the population industrial (USCENIND) and occupational (USCENOCC) classifications, which consist of 231 industrial and 503 occupational three-digit codes (16).
The NOES involved walk-through assessments carried out by trained engineers in a stratified sample of 4490 workplaces within the United States during 1981 to 1983. The number of workers exposed to over 12 000 chemical, physical, and biological agents employed within 121 industrial and 377 occupational categories was recorded. Exposure was defined as that persisting beyond 30 minutes per workweek or occurring at least once per week for 90% of the weeks of the work year. Some industries were excluded by NOES due to few expected exposures (eg, financial, insurance, and real estate facilities). We assumed these industries were unexposed.
Comparison of Canadian job exposures via the US NOES-JEM was facilitated by a linkage system previously developed by experts in occupational coding that translated the 860 CSIC codes to USCENIND equivalents and the 503 CSOC codes to USCENOCC equivalents (17). The relationship between the Canadian and US codes was often not one-to-one. For example, CSOC 8562 upholsterers translates to two USCENOCC codes (668 upholsterers and 785 assemblers). Eighty-four percent of the Canadian industries had exactly one US translation, and 16% had more than one US translation with a maximum of seven translations. Forty-two percent of the Canadian occupations had one US translation, and 58% had more than one US translation with a maximum of 31 translations. We considered each translation for a particular Canadian occupation or industry equally applicable. Thus the probability of exposure in a given Canadian job was obtained by first translating the Canadian industry and occupation codes into US equivalents and then averaging the exposure probabilities obtained from the NOES-JEM for all possible translated US industry and occupation combinations. We assumed exposure was zero for translated US industry-occupation combinations not found in the JEM.

Cumulative exposure
Cumulative exposure (expressed as expected work years of exposure) to each chemical agent was calculated as the product of the probability of exposure and job duration (in work years), summed across all recorded Canadian jobs. A work year was defined as 1 year in a full-time job or 2 years in a part-time job. All job histories were truncated at 5 years prior to diagnosis to account for possible latency periods.

Data analysis
The controls comprised internal controls of all other cancer sites, excluding lung cancer and cancers of an unknown primary site. Completed questionnaires were available for 1129 possible bladder cases, mostly transitional cell carcinoma (94%), and 10 628 eligible controls. The reliability of the information on occupational histories has been examined previously and was reported to be very good (9). A further 63 cases (5.6%) and 453 controls (4.3%) were excluded due to incomplete occupational histories. Controls were matched to the cases according to year of birth and year of diagnosis. Matching resulted in 1062 bladder cancer cases and 8057 cancer controls for the analysis. The median number of matched controls per case was 7.
We excluded the biological and physical agents and restricted our analysis to the remaining 12 456 chemical agents covered by the job-exposure matrix. The risk of developing bladder cancer due to cumulative exposure was analyzed using a conditional logistic regression (18,19). The risk was also adjusted for potentially important confounders that were identified by entering all possible confounders into a model and taking a backwards stepwise approach retaining all variables with a P-value of <0.2. The following variables were considered to be possible confounders: ethnic origin, marital status, education, questionnaire respondent, alcohol consumption, and cigarette smoking status (eg, ever versus never, years since quitting, cigarettes per day, smoking duration, pack-years).
As nonlinear associations between the exposures and risk were anticipated, the cumulative exposure variable for each chemical agent was transformed to an ordinal scale with four categories. The first category was for no exposure. The nonzero cumulative exposures were split into tertiles according to the exposed controls. The categories were labeled "unexposed", "low", "medium", and "high" exposure and were coded as scores of zero, one, two, and three to construct an ordinal variable. Adjusted odds ratios (OR) and corresponding confidence intervals (CI) were estimated for ever versus never exposure for the chemical agents with at least three cases ever exposed. OR values with a P-value of <0.05 (two-sided test) were considered statistically significant; hereafter "significant" means "statistical significance" for brevity.
Chemical agents with at least three exposed cases were selected to be analyzed in this report. There is always a trade-off for such a choice. Choosing those with a large number of exposed cases would lead to more-precise estimates with less chance of false positives; on the other hand, important associations with a smaller number of exposed cases would be missed. Using a criterion with a small number of exposed cases would allow for opportunities to identify all potential associations, even those with only a few exposed cases. However, this approach would lead to less-precise estimates and a higher chance for false positives. The choice of at least 3 exposed cases is often used in epidemiology studies for hypothesis-generating objectives. This is an aim of this study since not many studies have reported on risks based on chemical agents. Furthermore, in principle, users can use the information provided in the appendix (placed on the homepage of the Scandinavian Journal of Work Environment & Health) to derive results corresponding with any higher cut-off point for the number of exposed cases.
Dose-response relationships for chemical agents with at least nine exposed cases were examined. The OR values and corresponding confidence intervals were estimated for low, medium, and high versus zero exposure. A linear trend test was considered by treating the cumulative exposure on the transformed ordinal scale as a continuous variable.
As many chemical agents and hence hypotheses were tested, multiple comparisons were accounted for by controlling the false discovery rate (FDR) to below 5% according to Benjamini & Hochberg (20). This FDR method controls the level of false positives expected among the selected chemical agents; that is, a 5% false discovery rate means that it is expected that 95% of the selected chemical agents have been correctly identified. This method is less conservative than alternative methods that control the probability of at least one false positive, such as the Bonferroni method. The chosen method is intuitively more appealing in our study since it looks at the type-I error rate in the list of chemical agents selected (as opposed to the probability of any type-I error-commonly referred to as the P-value), and has greater power than other procedures.
Chemical agents that had at least nine cases exposed and that, after adjustment for multiple comparisons, either exhibited a significantly increased OR value for ever exposure or a significantly increasing linear trend test were considered as potentially exhibiting bladder carcinogenic properties. These chemical agents were selected for further analysis.
High correlations between the cumulative exposures of some of the selected chemical agents were expected, as similar levels of exposure with different agents could occur within the same jobs. A principal component analysis with Varimax rotation was performed on the ordinal cumulative exposure variables to examine the interrelations among the selected chemical agents (21). Chemical agents were assigned to the principal components with which they had component loadings greater than 0.4.
In addition, results are presented for chemical agents classified by IARC as definite or probable carcinogens for bladder cancer (6, 7).

Results
Matching resulted in 1062 bladder cancer cases and 8057 cancer controls for the analysis. The controls were diagnosed with cancer at a range of primary sites, the most common being prostate (17.4%), nonmelanoma skin (13.2%), and colon (13%). Cases and controls were comparable with regard to age, work years employed, education, and marital status (table 1). However, the bladder cases were more likely to be Caucasian, to have answered the questionnaire themselves, and to have smoked (table 2). These potentially important confounding variables were adjusted for in all of the analyses.
The study participants reported an average of 4.6 jobs and 36 work years. Distributions of the types of occupations and industries have been previously reported (9). All Canadian occupations and industries could be translated into US equivalents, except for jobs in the armed forces (4.6%), which were assumed to be unexposed to all chemical agents. Each reported Canadian job translated into an average of 3.5 US industry-occupation combinations. Fifty-two percent of the jobs in this study were not listed in the NOES-JEM. These include 25% of the jobs in industries that were excluded in the NOES survey (14, 15) either because they had low potential for any exposure (government 6%, education 2%, and finance, insurance and real estate 4%) or because their exposures were thought to be too heterogeneous (agriculture 8%, mining 4%, and fishing and trapping 1%). The remaining jobs were in surveyed industries but were not listed in the NOES JEM, possibly due to low or no exposure, including retail and wholesalers (10%), business and others services (6%), and transportation (3%). Overall, 74% of the study participants had some occupational exposure.

Occupational chemical agents
The ever exposure results for 5661 chemical agents with at least three exposed cases plus the additional dose-response results for 3419 chemical agents with at least  Chemical agents with three or more exposed cases. Although 283 chemical agents were expected to be significant by chance alone, a significantly increased OR for ever exposure was detected for 635 chemical agents, of which 163 had OR values above two. After adjustment for multiple comparisons, 7 chemical agents remained significant (table 3).
Chemical agents with nine or more exposed cases. Among the identified chemical agents, several are petroleum products or additives. Others include chemical agents used in the formulation of lubricant oils, engine oils and grease; those used in solvents or paints; and soaps and detergents (tables 3 and 4).

Principal component analysis
Performing a principal component analysis on the identified 29 chemical agents resulted in five principal components (table 5). The largest principal component accounted for 26% of the total variance, and the five principal components combined accounted for more than 75% of the total variance. The chemical agents tended to be grouped according to the jobs in which their exposures mainly occurred. Exposures comprising the first and second principal components were predominantly due to employment in the logging and gasoline service station industries, respectively. The third principal component mainly consisted of exposures from the construction and automotive repair shop industries. A considerable proportion of the exposures in the fourth principal component were due to employment as a truck driver. Members of the fourth and fifth principal components also had exposures in the gasoline service station industry. Table 6 presents the ever exposure and dose-response results for chemical agents identified by IARC (6, 7). Among the seven chemical agents with available data for evaluation, statistically significant or at least a 50% increased risk was observed for four chemical agents (mineral oils, benz(a)anthracene, 4-chloro-ortho-toluidine, and diesel engine exhaust) which also exhibited a significant dose-response relationship (P=0.01). Furthermore, a borderline significant increased risk was found for coal-tar pitches (OR 1.42, 95% CI 0.99-2.04). A nonsignificant 36% increased risk was found for direct black 38, a benzidine-based dye. No excess risk was observed for ortho-toluidine.

Discussion
Many characteristics of this study strengthen our findings. Both the cases and controls were selected from population-based cancer incidence patients, and all of the cancers were histologically confirmed. The large Table 3. Conditional logistic regression-adjusted a risk estimates for ever exposure to seven chemical agents with significant associations after adjustment for multiple testing in a bladder cancer study, British Columbia, Canada, 1983Canada, -1990  Richardson et al number of cases and controls enabled us to evaluate risk on the basis of exposure to a specific chemical agent rather than according to less precise employment categories. Applying a generic job-exposure matrix enabled us to study a range of specific chemical agents, usually prohibitive for studies utilizing expert assessments. The exposure estimates were also more reproducible and objective than those from the expert assessments (22). Over 80% of the patients completed the questionnaire themselves, and the questionnaire was found to be valid and reliable (10). Differential recall bias was minimized because cancer controls were used. Complete job histories were provided, enabling more precise measures of cumulative exposure through the incorporation of both estimated probability and the duration of exposure within all jobs. Besides adjusting for confounding factors, we controlled for multiple comparisons in the analysis to reduce the chance of false positives. A limitation of this study, common for studies utilizing a job-exposure matrix, was the potential for Table 4. Conditional logistic regression-adjusted a risk estimates for occupational exposure to 29 selected chemical agents with significant associations in a bladder cancer study, British Columbia, Canada, 1983Canada, -1990. [OR = odds ratio; 95% CI = 95% confidence interval; RP = reaction products; CIH = chlorinated isobutylene homopolymer; AE = 1-(2-((2-((2-((2-aminoethyl)amino)ethyl)amino)ethyl)amino)ethyl)-, monopolyisobutenyl derivatives; PW1 = polymer with 1,3-butadiene and ethenylbenzene; 2ME = 2-me-, c12 ester, poly w/ c16 2me2propenoate, iso-c10 2me2propenoate, me 2me2propenoate, c18 2me2propenoate, c14 2me2propenoate; S = sulfurized dodecylphenol calcium salt, sulfurized; NEC = not elsewhere classified; 2B = 2-  misclassifying exposure. It is, however, more likely to be nondifferential and may underestimate the true association; thus negative findings should be regarded as inconclusive (23). Assuming zero exposure for industries not studied by NOES may not be realistic since excess bladder cancer has been suspected in some of these industries, such as agricultural production and mining (1,5,24). However, these industries did not exhibit excess risk in this study population (9). Assuming no exposure for jobs in the armed forces may also be unrealistic for some exposures (eg, combustion products). However, no excess was observed for this group in an earlier analysis (9). Furthermore, the estimated OR values from the current study changed only slightly (<10%), with no changes in significant results, when patients ever in the armed forces were excluded. We averaged the exposure probabilities over all of the translated US industry-occupation combinations, as we did not have information on which translations were more likely. We also assumed the exposure was zero when a translated US industry-occupation combination was not in the job-exposure matrix, as we were unable to distinguish between it being invalid in the United States or having been observed to be unexposed. Thus cumulative exposures were subject to some measurement error from averaging over exposures for invalid or inappropriate  The compatibility of exposure levels between the US industries and Canadian or British Columbia equivalences is a concern. However a comparison of the jobexposure matrix specifically developed for the lead and zinc industry in British Columbia and that from the NOES-JEM showed good concordance overall (25).
We were also unable to account for variations in the concentration or frequency of exposure, variation within jobs, or changes across time. Although the NOES-JEM may not be very specific in comparison with expert assessments, self-reports, and other job-exposure matrices (26), our study found consistent results for definite or probable bladder carcinogens. There is also the potential for selection bias due to the use of cancer controls, a procedure which biases estimates towards unity (27). Lung cancer controls were excluded and a range of other cancer controls was used to limit this bias (19). Nonresponse bias has been examined earlier and found to be minimal (10). By various means, minimal information, including years of schooling, ever smoking, and usual occupation, was obtained for 227 nonrespondents. Comparisons with the respondents showed no significant differences, except for managerial occupations. Questionnaire reliability and recall bias were also examined, and excellent concordance was observed with interclass correlations in the range of 0.95 (10).
This case-control study showed a significant positive association between the risk of developing bladder cancer and exposure to several specific chemical agents, after adjustment for confounding factors and multiple testing (tables 3 and 4). The exposures were based on the NOES-JEM with specific chemical agents, and relatively few of them were studied in association with cancer. Most of the identified chemical agents can, however, be broadly grouped into subcategories as petroleum products or additives, lubricating oils and grease, paints and solvents, and soaps and detergents.
For petroleum products and additives, we observed a significantly elevated risk for asphalt and gasoline combustion, in addition to other specific chemicals that have, to our knowledge, not been reported previously in the literature. Asphalt (bitumen) has remained unclassifiable by IARC; however, it is on its high-priority list for reconsideration (6). A few other studies have reported a positive association (28,29). Gasoline and diesel engine exhaust are currently classified by IARC as probably and possibly carcinogenic, respectively (30). Several other studies have indicated an increased risk for exposure to petroleum and combustion products, suggesting an association with PAH (7,(31)(32)(33)(34). An earlier analysis of this case-control study on the basis of occupations and industries also suggested a link with PAH exposure (9) through employment in asphalt paving, truck driving, mechanic work, gasoline service station attending, printing, painting, and the metal and aluminum industries. IARC has classified some PAH as probably carcinogenic and many of the occupational circumstances with high PAH exposure as definitely or probably carcinogenic (35). are presented for chemical agents with at least three exposed cases. Data were not available for chemical agents that were not included in the NOES-JEM. When multiple NOES-JEM equivalents were found for a bladder carcinogen, the results for the one with the greatest number of exposed cases were given. b Code of the National Occupational Exposure Survey (NOES) in parentheses. c No NOES-JEM chemical agent equivalent could be found for 4-aminobiphenyl and dibenz(a,h)anthracene. Fewer than three exposed cases were found for benzidine, 2-naphthylamine, benzo[a]pyrene, 4,4'-methylene bis(2-chloroaniline), and benzidine-based dyes direct blue 6 and direct brown 95. d See the text for an explanation of the ever and dose-response exposure levels. The zero cumulative exposure group was always used as the reference category. e The P-value in italics represents a decreasing linear trend.

Male bladder cancer risk and occupational exposure
We found a significantly increased risk for several specific chemicals used in the formulation of engine oils, lubricating oils, or grease. Although these chemicals have generally never been studied individually in relation to cancer, a recent study from the four western Canadian provinces has found significant excess risk for exposure to lubricating, cutting, and mineral oils (36). Several past studies have also observed strong evidence for such an association (7,(37)(38)(39). It is interesting to note that a recent biological study has found that exposure to PAH or other compounds from skin contact with engine oils may cause oxidative DNA (deoxyribonucleic acid) damage (40), suggesting that a dermal absorption of carcinogens can lead to cancer development.
Several specific solvents, some used, for example, as components of polymer paints, have been observed to have a significant association with the risk of developing bladder cancer in this study. Past case-control and cohort studies have suggested an association with exposure to solvents and paints (5,(41)(42)(43). A recent study in the Netherlands found a nonsignificant excess for those exposed to paint components (33). However, only general components were used in these studies; therefore, information on their association with specific chemical agents was not available.
In this study, a few specific surfactants used in soaps and detergents were significantly associated with bladder cancer risk. These specific chemical agents have not been reported individually in the literature. However, an excess risk for exposure to soaps and detergents was observed in a mortality study (44).
Significant excess risk was observed for clay in this study. Exposure to clay dust was found to be associated with bladder cancer in a Montreal study (8). The fact that clay dust is a common exposure within the clay-, glass-, and stone-processing industries may explain why an excess risk for exposure to clay was found in this study.
This dataset was originally analyzed with the use of job title and industry. Within each job or industry, workers may possibly be exposed to several hundred chemical agents. Looking at exposure to individual chemical agents, as in this study, led us to a smaller subset of potentially important chemicals. However workers may be still exposed to several chemicals within this smaller subset concurrently since several chemicals may be present together in a given job. The principal component analysis was thus used to identify subgroups of chemical agents (commonly called components) for which workers tended to be exposed together within this subset. For each subgroup, the results provided chemical agents that may individually or collectively contribute to an increased risk of bladder cancer. Due to high correlations, it is generally not possible to determine whether the chemical agents within each subgroup are individually or synergistically responsible for the observed association. However, the results could be used to suggest possible causes for increased risk on the basis of common characteristics of chemical agents within each subgroup.
The results from the principal components analysis suggest that the risk from the first component could be due to exposure to PAH (table 5). Many of the constituents are comprised of PAH or involve potential PAH exposure, such as asphalt and products of the combustion of gasoline (leaded and unleaded) and jet fuel. The remaining constituents of this component are ingredients for various engine, compressor, and lubricating oils, which could also have carcinogenic properties due to certain mineral and fuel oils being classified by IARC as definitely or possibly carcinogenic, respectively (6).
Exposures contributing to the second component were mainly due to employment in gasoline service stations. Excess bladder risk has been noted for this industry in this population and elsewhere (1,9). Various engine emissions, diesel or traffic fumes, PAH, lubricating oils, and greases could be responsible for the excess risk. Only service station attendants had possible exposure to liquefied natural gas, and, as such, were potentially exposed to natural gas combustion products, which were noted as potential carcinogens in a Montreal study (8). Ether, tert-butyl methyl is still unclassified by IARC (45) due to inadequate evidence with respect to human exposure; however, there is need for further research into this chemical agent as a potential carcinogen (46).
The construction and automotive repair industries contributed much of the exposures in the third component. Excess risk could be due to exposure to oils, solvents, or metal dusts. Clay and various metals have been suggested as bladder carcinogens (8,9). A positive association of marginal significance between tin exposure and bladder cancer was also observed in a German case-control study (47).
There are no clear carcinogens in the remaining two components, but many of the chemical agents are again oils or products of petroleum. A significant proportion of the exposures in the fourth component were due to employment as a truck driver, which has been highlighted as at excess risk in many studies (47)(48)(49).
Results from our study are fairly consistent for definite or probable bladder carcinogens as classified by IARC (table 6). Of the specific PAH, we observed a large and significant excess risk for ever exposure to benz(a)anthracene and a borderline significant association for coal-tar pitch. Coal tar is a known carcinogen and a probable contributor of excess risk in the aluminum industry (1,6,50). It should be noted that benzidine-based dyes and ortho-toluidine can occur in painting and printing, and they contain aromatic amines. They are both classified as probable carcinogens by IARC (6,51). Benzidine-based dyes exhibited excess Richardson et al risk, although not at a statistically significant level, and ortho-toluidine did not exhibit any major excess risk in our study.
This study also replicated known associations between bladder cancer and Caucasian ethnicity and cigarette smoking (1). Smoking duration had a strong association with bladder cancer; yet it was thought to have a minimal confounding effect on the exposure results of the chemical agents. We examined the risk of exposure to the 29 selected chemical agents without adjusting for smoking duration, and all but two had a <10% change in their OR values. The larger changes for the two chemical agents may be due to the smaller number of cases (N=25) and hence to a greater variation.
In conclusion, in our study, the use of a generic job-exposure matrix to estimate cumulative exposure corroborated important findings from the literature and documented bladder cancer risk for specific chemical agents not previously reported.