Identification of cases and controls

Multiple myeloma cases were identified through the Surveillance, Epidemiology, and End Results (SEER) cancer registries in the Seattle-Puget Sound region of Washington and the Detroit metropolitan area of Michigan. These registries identified cases by reviewing medical records at all hospitals in the region, as well as by reviewing the records of selected pathology laboratories, oncologists, radiologists and state death certificates. The frequency of these reviews enabled the identification of cases within 1–2 months of diagnosis. Cases eligible to participate in this study were 35–74 years old and were diagnosed with multiple myeloma (ICD-O-2/3 9731:9732) between 1 January 2000 and 31 March 2002. Of the 365 cases who appeared eligible to participate in this study, 64 (18%) died before they could be contacted, 28 (8%) were unable to be located and the physicians of 18 (5%) refused. Of the remaining 255 who were contacted, 74 (29%) refused to participate, leaving a total of 181 cases. Thus, of cases who were alive, able to be located and whose eligibility could be confirmed, 71% (181/255) participated. Case participation was not associated with study site, age or gender. General population controls were selected from a case–control study of NHL that was undertaken at the same time and included the same geographical areas as used for multiple myeloma case recruitment.33 Controls under 65 years of age were identified using random digit dialling; controls 65–74 years of age were identified from the Center for Medicare and Medicaid Services files. We selected controls for our study from among the NHL study controls who (1) had not been previously diagnosed with multiple myeloma, plasmacytoma, NHL or HIV (the latter were excluded because of the association between the virus and lymphoma development),34 (2) were between 35 and 74 years of age, (3) were identified as residents of the Detroit or Seattle-Puget Sound areas between September 1998 and December 2002, and (4) spoke English. Of the 1133 potential controls who were eligible to participate in this multiple myeloma study, 52% (481/1133) participated as NHL study controls. Control participation was not associated with study site or gender, but individuals in the youngest (35–50) and oldest (65–74) age groups were less likely to have participated than those in the middle age group. Informed consent was obtained from each participant before the interview and approval was obtained from the appropriate Institutional Review Boards.

Exposure assessment

In-person interviews were conducted using a computer-assisted personal interview program. Surrogates were not permitted to complete interviews on behalf of study participants but were allowed to aid in recalling occupational histories, dates and frequencies of exposures. Information on all jobs held for at least 1 year between 1941 for cases and 1946 for controls and the dates of enrolment in the study was collected. Participants reported job start and stop years, job titles and employers for every job they held since age 18. The participants were also asked to describe what the employer made or what service was provided, their main duties, and how many months per year and hours per week they worked in each job. Additionally, for 20 occupations with potential solvent exposure, job-specific questionnaires developed by an industrial hygienist were administered.35 These included questions on the work environment (eg, ventilation practices), job-specific tasks performed by the subject, chemicals used and other factors relevant to exposure assessment.36 To minimise interview time, job-specific modules were administered only when participants held the relevant job for at least 2 years.

All jobs were coded according to the Standard Occupational Classification (SOC) system.37 In addition, all jobs held by each participant, regardless of whether job-specific questionnaires were used, were assessed for exposure to each of the six chlorinated solvents; a total of 2264 jobs were reviewed. Exposure metrics of probability, frequency, intensity and confidence were assigned by modifying job–exposure matrices (JEMs) based on the subjects' answers to the work history and the job module sections of the questionnaire. The JEMs were developed by an industrial hygienist after reviewing over 600 published papers and reports on chlorinated solvents (eg, see Bakke et al38 and Gold et al39). These JEMs were developed for each decade for specific (1) industries, such as the chemical or rubber industry, (2) occupations, such as auto mechanics or hair dressers and (3) tasks, such as degreasing, gluing and painting.

Exposure probability was defined as the theoretical percentage of workers reporting the same information who would have been likely to have had exposure to the solvent. Probability was scored as: 0=<1% of subjects were likely to have had exposure; 1=1% to <10% of subjects were likely to have had exposure; 2=10% to <50% of subjects were likely to have had exposure; 3=50% to <90% of subjects were likely to have had exposure; and, 4=self-reported use or ≥90% of subjects were likely to have had exposure. For jobs with probability scores of at least 1, frequency and intensity scores were also assigned. Frequency was defined as the average hours per week of exposure to a particular solvent, averaged over the job, and was categorised as: 0=<15 min/week; 1=15 min to <1 h/week; 2=1 to 10 h/week; 3=>10 to 20 h/week; and 4=>20 h/week. The intensity score was the concentration of solvent estimated to have been in the subject's breathing zone, in parts per million (ppm), during the exposure period (not necessarily for a full shift) and was coded as: 1=<1 to 10 ppm; 2=>10 to 100 ppm; 3=>100 to 200 ppm; and 4=>200 ppm. Finally, the confidence level was assigned as: 1=literature contradictory or no information was available; 2=one metric (probability, frequency or intensity) was based on the literature or self-report; 3=two metrics were based on the literature or self-report; and, 4=all three metrics were based on the literature or directly from the subject's report.

Some general rules were applied across all jobs. If the subject reported information that contradicted the appropriate JEM scores, the subject-specific data were used. If chlorinated solvent exposure was not specifically mentioned but would normally have occurred when performing reported tasks (eg, an auto mechanic reported doing tune-ups and clutch or brake work, but did not report degreasing), the subject was assessed as having been exposed and given probability, frequency and intensity scores from the JEM, but the confidence score was lowered by 1. If a subject reported working in an industry that likely involved tasks entailing chlorinated solvent exposure (eg, auto manufacturing) and the job title indicated s/he was not likely to do hands-on work in the production area, the frequency and intensity scores from the JEMs were lowered by 1. If dermal exposure was likely (eg, the subject reported degreasing with a rag), the intensity score was raised by 1. All assignments were conducted by an occupational epidemiologist (LSG) and reviewed by an industrial hygienist (PAS); both were blinded to case–control status.

We did not have information about non-occupational sources of chlorinated solvent exposure, such as use of glues or paints for hobbies, so we were not able to include these in each subject's cumulative solvent exposure. However, we did not expect these to be major sources of chlorinated solvent exposure relative to occupational exposures.

Statistical analysis

We used unconditional logistic regression to estimate ORs and 95% CIs for associations between the risk of multiple myeloma and having ever been exposed (defined as a probability score of at least 2 in any job) to any of the six chlorinated solvents and to each of the six chlorinated solvents. We also analysed whether the risk of multiple myeloma varied by duration, using the summed total years worked in jobs considered exposed to each chlorinated solvent. We performed analyses examining cumulative exposure over each subject's lifetime (for each exposed job, the midpoint of intensity (in ppm)×the midpoint of frequency (in hours/week)×total years worked, summed over all exposed jobs), again for jobs with a probability score of 2 or greater for a particular solvent. Finally, assuming multiple myeloma develops over a number of years and recent exposures are less relevant than those that took place in the more distant past, we conducted analyses in which potential exposures that occurred within 10 years of the reference date were considered unexposed. Each of the continuous exposure metrics (duration, cumulative and lagged cumulative exposure) was categorised into four groups according to quartiles of the control exposure distribution. Tests of trend were conducted using a linear term for the median duration and cumulative scores among controls in each category. The reference group in each model consisted of study participants who were assessed to have never been exposed (probability <2 in all jobs) to that particular chlorinated solvent.

All models included the covariates gender, age (35–50 years (referent), 51–64 years and 65–74 years), race (only white (referent), any black, any Asian and other), education (less than 12 years (referent), 12–15 years and 16 or more years) and SEER site (Seattle and Detroit). Reported p values are two-sided and p values of <0.05 were considered to be statistically significant.

As a sensitivity analysis, we considered all occupations that were given confidence scores of 1 as unexposed and repeated all analyses. Evidence has shown that the main source of non-differential misclassification that falsely attenuates ORs towards the null is incorrect classification of unexposed individuals as exposed in population-based case–control studies in which the prevalence of exposure is low. Incorrectly classifying an equal proportion of exposed individuals as unexposed has been reported not to have substantial effects on risk estimates.40–42 We also performed an additional sub-analysis with stratification by SEER site to evaluate whether the different case-to-control ratios in Seattle and Detroit influenced our results because a greater proportion of controls than cases were from Seattle and the prevalence of industries, occupations and chlorinated solvent exposures also differed by study site. For the site-stratified results, we note associations that were present in only one site. Finally, because uses of TCA, TCE, DCM and carbon tetrachloride may have been correlated, we report the percentages of control subjects exposed to these chemicals alone and to two of these chemicals and provide an estimate of the association with multiple myeloma for subjects who were exposed to all four.