The general methodology of the study has been described in detail in a previous article [30]. Briefly, male cancer patients aged 20 years and older ascertained by the population-based British Columbia Cancer Registry (BCCR) for the years 1983 to 1990 inclusively, received a self-administered questionnaire requesting lifetime job descriptions, occupation and industry titles, and duration and period of work, as well as information on ethnic origin, education, and lifetime smoking habits. Questions about lifetime consumption of alcoholic beverages, which were initially omitted, were added to the questionnaire during the first year of the study. In the event of a patient’s death, information was requested from the spouse or closest living relative. For the first two years, questionnaires were sent to all cases. Data collection continued until 1,000 completed questionnaires were accrued for each tumor site, or until 31 December 1990, whichever occurred first. The distribution of patients’ usual occupations and usual industries as well as assessments of response bias and of questionnaire validation and reliability was conducted and previously reported [30]. The anatomic site and pathology of the primary tumor were coded using the 9th revision of the International Classification of Diseases [31] and the International Classification of Diseases for Oncology [32], respectively. Occupations and industries were coded according to the Canadian Standard Occupational Classification (SOC) and the Canadian Standard Industrial Classification (SIC), respectively [33,34]. In the SOC, occupations are coded according to two-digit major group codes, three-digit minor group codes, and four-digit unit group codes. In the SIC, industries are coded according to two-digit major group codes, three-digit group codes, and four-digit class codes. During the data-collection period, questionnaires were sent to all 25,726 eligible male cancer cases ascertained by the BCCR and 15,463 (60.1%) were returned. Histological confirmation of diagnosis was obtained in all cases. For colon cancer, questionnaires were sent to 1,768 eligible cases, of which 1,156 (65.4%) were returned.

A group matched case-control design was used. Cases comprised all 1,156 colon cancer patients. Controls comprised internal controls with all other cancer sites, excluding lung cancer (n = 2,998), rectum cancer (n = 1,095) and cancers of unknown primary site (n = 708), matched to the cases on exact age and year of diagnosis. These were based on a few methodological considerations. First, using other cancers as controls has a few advantages over population controls with regard to recall and interviewer bias when using general population controls, a mixture of live and deceased controls can be expected when selecting other cancer patients as controls and there are logistical and cost advantages in using other cancer patients as controls. Thus the study design using cancers as controls is likely to be used increasingly in the future, particularly in cancer registry settings [35]. Next, since lung cancer is mainly caused by smoking, excluding it from the control group would avoid potential bias of having too many smokers in controls. In addition, matching on age and year of diagnosis is a way to control for major confounders. Including year of diagnosis is a control of time when rapid changes in population structure were happened from 1960s through 1990s.

Data were analyzed using conditional logistic regression for matched sets and the likelihood ratio test [36,37] in a two-step procedure. In step 1, the effect of the following non-occupational confounding variables was assessed: marital status; education (less than 8 years; 8 to 11 years; high school; post-secondary); smoking (starting age at smoking, average number of cigarettes, pipe or cigars smoked per day, total years smoked); alcohol consumption (starting age at consuming alcohol; average of number of bottles of beer, glasses of wine, ounces of spirits consumed per day; total years of consuming alcohol); and person who filled out the questionnaire (self or proxy). Variables were selected in a forward fashion each being examined separately; Potentially important confounders (P < 0.2) were then included in the model and the remaining ones were examined; the process was repeated until no factor with P < 0.2 was identified. In step 2, each occupation and industry was assessed separately using conditional logistic regression in which all significant confounding variables identified in step 1 were taken into account. The cut point of 0.20 was chosen in step 1 to ensure that the process has sufficient power to identify potentially important confounders.

Matched case-control analyses were carried out using PECAN software [38], test of significance of the adjusted odds ratios (ORs), p-values and 95% confidence intervals were calculated and presented. Analyses were performed for the following two different estimates of occupational exposures: ever occupation/industry (whether a job was ever held in a given occupation/industry under consideration) versus never, and usual occupation/industry (job with the longest held lifetime employment in a given occupation/industry). People in either ever or usual occupation/industry categories were always compared to their counterparts who were never in such occupation/industry. Due to all 597 occupation titles and 1,104 industry titles were analyzed and small numbers of cases in many occupation/industry categories, we intended to maximally present results of the analyses in this article with occupations and industries as long as their number of cases is five and over.

In the usual occupation category, significant excess risk in the major occupation groups (two-digit codes) was only observed at P = 0.10 for occupations in social sciences and related (SOC 23). Excess risks for minor groups (three-digit codes) were significant at P = 0.05 for occupations in fabricating, assembling and repairing wood products (SOC 854) and at P = 0.10 for occupations in other managers and administrators (SOC 113/114), bookkeeping and account-recording (SOC 413), and lodging and other accommodation services (SOC 613). Several unit occupations (four-digit codes) had significantly increased ORs at P = 0.05 for: insurance sales jobs (SOC 5171), rail transport equipment mechanics and repairers (SOC 8583), brick and stone masons-tile setters (SOC 8782), ship engineering officers (SOC 9153), and at P = 0.10 for: administrators in teaching and related fields (SOC 1133), financial management (SOC 1135), livestock farm workers (SOC 7183), sawmill sawyers and related (SOC 8231), cabinet and wood furniture makers (SOC 8541), and taxi drivers and chauffeurs (SOC 9173); the ORs were significantly low at P = 0.10 for sales clerks and salespersons in commodities, not else classified (SOC 5135) and welding and flame cutting (SOC 8335).

In the usual industry category, a significant excess risk in the major industry group was noted at P = 0.05 for motor vehicle, parts and accessories, wholesale (SIC 55) and insurance and real estate agencies (SIC 76) and at P = 0.10 for beverage (SIC 11), federal government services (SIC 81) and membership organizations (SIC 98); ORs for printing, publishing and allied (SIC 28) industry was low at P = 0.10. Excess risks were significant at P = 0.05 for the following minor industry groups (three-digit codes): brewery products (SIC 113), defense services (SIC 811), and human resource administration (SIC 826) and at P = 0.10 for livestock farms (except animal specialties) (SIC 011), and dairy products (SIC 104). Odds ratio was significantly low at P = 0.10 for interior and finishing work (SIC 427). Several industry classes (four-digit codes) had significantly increased ORs at P = 0.05 for poultry and egg farms (SIC 0114), construction, mining machinery, materials handling (SIC 3192), exterior close-in: masonry work (SIC 4231), used goods moving and storage (SIC 4562), and specialty food stores (SIC 6012). ORs were significantly low at P = 0.10 for general freight trucking (SIC 4561), motor vehicle repair garages (general repairs) (SIC 6351) and general hospitals (SIC 8611).

In the validation study of the questionnaire, company job records of 81 individuals who reported having been employed in one of two large companies in BC were compared with information from their questionnaire; and the interclass correlation was 0.996 for starting year of employment and 0.971 for duration of employment.