Activities and Exposures during Leisure and Prostate Cancer Risk¹

International variation in incidence rates of clinically evident prostate cancer suggests that its development is related to exogenous factors (1). However, the only well-established risk factors are age, family history of the disease, race, and country of residence (2).

Numerous studies of prostate cancer have attempted to identify causally related occupational exposures. Ross and Schottenfeld (3), however, pointed out that if any of the implicated exposures are causes, they would account for a very small proportion of cases, because few men were exposed in relevant industries. One class of exposure circumstances that has not been investigated is that of hobbies and leisure activities. Some men spend considerable time involved in activities such as gardening, woodwork, and automobile maintenance, often with little protection from chemical exposures. We examined the effects of activities and exposures during leisure on prostate cancer risk using data from a case-control study of cancer at multiple sites, including prostate cancer, in Montreal. The study was designed primarily to assess occupational exposures, but included information on activities and exposures during leisure (4, 5).

Details of the overall study design were described elsewhere (4, 5). Between 1979 and 1985, almost all (97%) histologically confirmed incident cancer cases at 21 sites occurring in men aged 35–70 years that were diagnosed at all of the large hospitals in metropolitan Montreal were accrued. The study was approved by the ethics committees of all participating institutions.

Of 557 eligible cases of prostate cancer [International Classification of Diseases, version 9 (ICD-9), 185], data were obtained from 449 (80.6%) either by face-to-face interviews of the patients or proxies or by a shorter self-administered questionnaire given to those unwilling to be interviewed. The case series was restricted to the 400 prostate cancer cases who underwent the face-to-face interviews (age range, 47–70 years). To assess possible nonresponse bias, information on age, income, ethnicity, marital status, cigarette smoking, and alcohol consumption was abstracted from the medical records of all prostate cancer cases and from a comparison group consisting of other cancer patients (6). By comparing ORs³ expressing the associations between these attributes and prostate cancer among all subjects (the respondents and the nonrespondents) to similar ORs for the subjects who underwent the face-to-face interviews (the respondents), Richardson (6) showed that there was virtually no nonresponse bias in relation to these attributes.

During the same period, 740 population controls were selected using as sampling frames electoral lists or random digit dialing. Electoral lists are compiled in Canada by active enumeration of citizens and are adequately up to date for about 2 years after an election. Thus, we used the lists of all Montreal area electoral districts within 2 years of enumeration to select a random sample of men, frequency matched by age and area of residence to the cancer cases (all sites combined). Some 541 controls were selected of whom 375 (69.3%) provided information (4). During 1983, we selected controls by random digit dialing. Stratifying on the first three digits, four digits were randomly selected, and the number was dialed. Seven callbacks were made for nonrespondent numbers before abandoning them. When someone answered, we asked if it was a commercial or residential number and, if the latter, if there was a man in the appropriate age stratum residing there. Once the quota of men was obtained, we sent each an introductory letter. Some 199 controls were selected by random digit dialing; 158 (79.4%) provided information. In total, 533 of the 740 controls selected provided information (72%; Ref. 4). Of these, 512 underwent the face-to-face interviews. We used data from the 476 interviewed controls who were between 45 and 70 years of age on the assigned date of pseudo-diagnosis, on the basis of the interview year.

The interviews used to obtain data were on the basis of a questionnaire and were carried out by a team of three interviewers, selected on the basis of their outgoing personalities, nonjudgmental approach, ability to speak both English and French, and experience in the workplace (4, 5). Because of turnover, there were six interviewers during the 6 years of fieldwork. Of three women, two were nurses, and one was a secretary; of three men, one was a machinist, one a salesman, and one a biology teacher. They were trained in the administration of the questionnaire for 1 week using role playing and videotaped feedback.

Some prostate cancer cases were interviewed in hospitals and others at home. All controls were interviewed at home. It was impossible to blind the interviewers to subjects’ disease status, though usually they were blinded to the cases’ type of cancer.

The questionnaire consisted of two sections: (a) a structured section addressing nonoccupational exposures; and (b) a semi-structured section focusing on the occupational history. During the structured section, subjects were asked if they spent time engaged in a variety of hobby and leisure activities, presented as a checklist, and if any of these activities involved exposure to 12 substances. For each activity and each substance, the respondent was asked if they were ever exposed regularly, defined as once a week or more for ≥6 months; otherwise, they were considered unexposed. No further information was collected on details of these exposures, such as their frequency, duration, and timing during the subjects’ lives. Thus, each exposure was considered as a simple dichotomy.

The occupational history led to an in-depth exposure assessment using a checklist of 294 substances, which included the substances on the hobbies checklist. Thus, for the latter substances, we had information about exposure from two different sources: occupational and leisure. A team of chemists and industrial hygienists performed the occupational exposure assessment. They examined each employment history and translated each job into a list of potential chemical exposures, after reaching a consensus, using the checklist that included 294 substances (4, 5). Chemical coding was carried out blindly with respect to the subjects’ disease status. For each product considered present in each job, the coders noted three dimensions of information, each on a three point scale: (a) degree of confidence that the exposure had occurred (possible, probable, or definite); (b) frequency of exposure in a working week (<5%, 5–30%, or >30% of the time); and (c) concentration to which the worker was exposed (low, medium, or high). In analyses, we dichotomized each attribution of exposure as being “substantial” or “not substantial” using an algorithm that accounted for the degree and duration of exposure (7).

During the interview, subjects were also asked if they smoked cigarettes nearly every day. If so, they were asked the age at which smoking began, the age at which they stopped (for ex-smokers), and the average daily amount smoked. We translated intensity and duration of cigarette smoking into a cumulative exposure variable, defined as the product of the average number of cigarettes smoked per day and the duration of smoking in years and expressed as “pack-years” (20 cigarettes/pack). When values of components of this cumulative variable were missing, they were imputed with median values.

Similarly, we constructed a composite exposure variable representing cumulative exposure to alcohol by summing the cumulative exposure variables for each beverage (beer, spirits, or wine). If values of components of the cumulative variables were missing, they were imputed with median values. When the study was carried out, the ethanol content of the usual portions of each beverage was estimated as 13.6 g (8).

BMI (weight/height²) was calculated using subjects’ usual weight when in good health.

When dividing distributions of covariates into tertiles, cutpoints were determined from the distributions among the population controls.

We calculated ORs between prostate cancer and activities and exposures during leisure to estimate incidence rate ratios, while adjusting for age, ethnicity, respondent status, family income, BMI, cumulative cigarette smoking, and cumulative alcohol consumption with unconditional logistic regression (9) using categorical variables, fitted with the SAS LOGIST procedure (10).

The mean age of the cases (n = 400) was 62.9 years (SD, 5). For the controls (n = 476), it was 60.3 years (SD, 6.6). Table 1 shows other characteristics of the subjects, who were mainly of French origin. Most of those who were not French were of other European origins.

The mean age of the 49 cases excluded because they did not undergo the face-to-face interviews [63.5 years (SD, 4.5)] did not differ significantly from that of those included (P = 0.58), nor did their distribution of ethnic origins (P = 0.09). Those excluded were more likely to have had their data provided by proxy respondents (36.7%; P = 0.001) and to have had a lower annual family income (P = 0.002) and less likely to have been heavy smokers (P = 0.001). We had no data on their BMI and alcohol intake.

The mean age of the 57 controls excluded [53.8 years (SD, 13.6)], because they did not undergo the face-to-face interviews (n = 21) or were aged <45 years or >70 years (n = 36), was significantly less than that of those included (P = 0.003). The distributions of ethnic origins (P = 0.68), respondent status (P = 0.36), annual family income (P = 0.85), BMI (P = 0.49), and alcohol intake (P = 0.23) did not differ significantly between those not included in the analysis and those who were. Those excluded were less likely to have been heavy smokers (P = 0.06).

Table 2 shows ORs for prostate cancer according to reported leisure activities. The ORs were either crude or adjusted for the covariates shown in Table 1, as well as age. Home or furniture maintenance was associated with an increased risk, as was painting, stripping, or varnishing furniture, whereas carpentry was not. Sports or other outdoor activities appeared weakly protective.

Table 3 shows crude and adjusted ORs for prostate cancer according to self-reported exposures during leisure. Of the 12 ORs estimated, 3 of the adjusted ORs were significantly elevated. Exposure to metal dust was associated with a significantly increased OR (3.2; 95% CI, 1.0–9.9), as was exposure to lubricating oils or greases (OR, 2.2; 95% CI, 1.2–3.7) and exposure to pesticides or garden sprays (OR, 2.3; 95% CI, 1.3–4.2).

Because we also had information on our subjects’ occupational exposure to these substances (7), we decided to examine risks in relation to different combinations of work and leisure exposure. We dichotomized work exposures as substantial versus not substantial; subjects in the latter category were considered unexposed at work. In Table 4, the referents consisted of subjects unexposed both at work and during leisure. For each of the three types of substance showing significantly increased risks in Table 3, the ORs associated with exposure during leisure without substantial exposure at work were greater than the ORs associated with substantial exposure at work without exposure during leisure. For lubricating oils or grease, the OR associated with exposure during both leisure and work was greater than the OR associated with exposure during either alone, whereas for pesticides or garden sprays it was not. The corresponding OR for metal dust could not be estimated because there were no controls with exposure during both leisure and work.

Our study had some particular strengths and potential limitations. The methods of accrual ensured that the case and control series were population based. Although selection bias could have occurred because of nonresponse, it was not appreciable in relation to the accrual of the cases with respect to age, sociodemographic attributes, and the use of cigarettes and alcohol (6). However, it could have arisen in the accrual of the controls because of nonresponse or by our restricting our analyses to those who underwent the face-to-face interviews.

We doubt that recall bias was appreciable. Most of the ORs shown in Tables 2 and 3 were consistent with no association. Whereas we found significant associations for a few substances, exposure to most of the substances was not reported more frequently among cases than among controls. Furthermore, those which were associated with prostate cancer were substances for which associations with occupational exposure were reported previously (7).

We adjusted for age. Confounding by race was unlikely, because almost all subjects were of European origin. Our adjusted ORs differed only slightly from the crude ORs, indicating that there was minimal confounding by the covariates. We had no information about family history. Nevertheless, because only a small fraction of prostate cancer cases can be attributed to known determinants (2), residual confounding by unknown determinants remains a possibility.

A generic problem in case-control studies of prostate cancer derives from the relatively high prevalence of occult prostate tumors among men >60 years of age. Without screening potential controls for occult tumors, it is impossible to ensure that a control group is free of prostate tumors. It is likely that some controls in our study had occult prostate tumors; this would have attenuated the true relative risks (11).

The information collected about subjects’ leisure activities was limited, and the dichotomization of exposure status was simplistic. Furthermore, we could not assess the validity of the reported leisure activities and exposures. Nevertheless, such misclassification would have attenuated true relative risks. There is no reliable evidence on the exposure levels incurred by hobbyists, and the range may be great. One should not assume that exposure levels are lower during leisure activities than during work. Anecdotal evidence suggests that some hobbyists experience very high exposure levels.

The association between home or furniture maintenance and prostate cancer (OR, 1.4; 95% CI, 1.0–1.9) cannot be explained by contact with wood or wood dust, because the ORs for carpentry and for exposure to wood dust indicated no excess risk. Similarly, exposures to paints, lacquers, stains, aerosols, spray paints, chemical solvents, dyes, plastic cement, and plastic resins cannot explain this association.

Another exposure encountered in furniture maintenance might account for part of the association, because the OR for painting, stripping, or varnishing furniture was elevated (OR, 2.1; 95% CI, 0.7–6.7). Paint strippers contain dichloromethane, for which there is experimental evidence of carcinogenicity in animals; inhalation produced increases in the incidence of benign and malignant lung and liver tumors in mice and rats (12). Dichloromethane is “possibly carcinogenic to humans” (12). Gibbs et al.(13) carried out a cohort mortality study of factory workers exposed to dichloromethane and found elevated SMRs for prostate cancer that increased with increasing exposure. For workers with high exposure, the SMR was 1.8 (95% CI, 1.0–3.1), and for workers used for ≥20 years, the SMR was 2.9 (P ≤ 0.05).

Exposure to epoxy resins may be involved in prostate carcinogenesis, because the OR associated with this exposure was 1.8, although the 95% CI was wide (0.6–5.7). Nonoccupational exposure to epoxy resins might occur during home maintenance involving the use of adhesives and contact with electrical insulation (4). Although cutaneous exposure to epoxy resins did not produce an increased incidence of tumors in mice (14), there is little evidence from epidemiological studies (15, 16).

We observed significantly elevated ORs, all >2, associated with exposure during leisure to metal dust, lubricating oils or grease, and pesticides or garden sprays (Table 3). Occupational exposure to these substances was shown previously to be associated with prostate cancer in our study, though the analyses were on the basis of a somewhat different configuration of cases and controls and confounders (7).

The dichotomizations of leisure and work exposures were not strictly comparable (Table 4). Thus, we cannot use the magnitude of the OR estimates to infer if work or leisure exposure induced a greater risk. Still, it is noteworthy that for the three substances evaluated, the ORs associated with leisure exposure were at least as large, if not larger, than those associated with work exposure. Given the small numbers in some cells, chance may account for this pattern. It could also reflect true risks under the hypothesis that subjects exposed during leisure take fewer precautions and are exposed at higher levels than workers.

The fact that excess risks were detected for both leisure exposure and work exposure to these three substances, and that the exposure data for the two settings were obtained independently of one another using different methods, adds to the plausibility of the associations. Data on the exposures during leisure were obtained from self reports elicited in a structured interview, whereas data on the occupational exposures were obtained by a team of chemists and industrial hygienists from detailed descriptions of subjects’ job histories elicited in a semi-structured phase of the interview.

Van der Gulden (17) reviewed 23 studies reporting incidence rate ratios for prostate cancer in relation to metal work, 13 of which exceeded 1; 2 were significant, with values of 1.8 and 2.2. He also reviewed 30 studies reporting estimates of mortality rate ratios, 20 of which exceeded 1; 1 was significant with a value of 1.8. Metal work may involve exposure to many different substances, including solvents, paints, lubricants, cutting oils, abrasives, and exhaust fumes (17). A recent review of the evidence on the association between metalworking fluids and prostate cancer considered it equivocal (18). Future studies should account for the potential confounding effects of a wider variety of associated exposures and identify the types of metal involved in carcinogenesis.

Although lubricating oils or greases can be of animal, vegetable, or mineral origin, most of the work exposures recorded in our study were of mineral origin, among motor vehicle repairmen, machinists, and farmers (4). Van der Gulden et al.(19) reviewed 11 studies of prostate cancer incidence among mechanics, repairmen, and machine operators; 9 reported rate ratios between 1.3 and 4.7. However, another study of theirs did not detect any increased risk associated with exposure to lubricating oils (20). Exposure to lubricating oils or grease has been associated with the development of stomach cancer (21), cancer of the supraglottis (22), and squamous cell skin carcinoma (23). The substantial increase in the risk of prostate cancer that we found associated with exposure during both leisure and work suggests that exposure to lubricating oils and grease warrants further investigation. Because the carcinogenic activity of lubricating oils can be eliminated by modifying their processing (24), their composition should be carefully specified (25).

We observed an increased risk attributable to exposure to pesticides or garden sprays during leisure. This is consistent with nine studies that have reported excess prostate cancer among farm workers or pesticide applicators (26, 27). Because some pesticides either mimic or antagonize the effects of sex steroid hormones, it is plausible that exposure to pesticides might increase prostate cancer risk by increasing cell proliferation (26).

We identified some exposures during leisure that are associated with the development of prostate cancer and for which there is some internal consistency and external corroboration. These associations require additional investigation to confirm our findings and to provide greater inferential specificity. The latter requires better information on exposure patterns and levels among subjects engaged in leisure activities and on protective measures.

We thank Dr. Lousie Nadon for helpful discussions and Marie Désy for assistance with data management.