Earlier epidemiologic studies have shown that occupational exposures may be associated with an increased risk of NHL. Increased risks of NHL were observed for workers in various industries and occupations, such as agriculture (8–12), metal production (10, 13), food (10, 14), pesticide applicators (15), painters and printers (13, 16), funeral directors and embalmers (13, 17), plumbers (13), dry cleaners (13), engineers, mechanics, and leather workers (10, 14). Occupational exposure to solvents has been suggested to be at least part of the reason for the observed increased risk; the data, however, linking solvent exposure to NHL risk have been inconsistent (13, 18–21).

One reason for the inconsistent results could be misclassification of exposure in early epidemiologic studies. Job titles have been widely used as surrogate measurements for occupational exposure to solvents. Under the same job title, however, workers may have very different exposure experiences or exposure levels for various chemicals. Use of job titles as surrogates for specific occupational solvent exposures could result in at least nondifferential exposure misclassification and thus underestimate the underlying associations between occupational solvent exposure and NHL risk.

In this population-based case-control study of NHL in Connecticut women, we used a job-exposure matrix to assess solvent exposure instead of relying on only job titles as surrogate measurements for the exposures of interest. The job-exposure matrix developed by industrial hygienists at the National Cancer Institute allows for semiquantitative measurements of the association between occupational exposures and NHL risk.

A standardized, structured questionnaire was used to collect information on lifetime occupational history and other major suspected risk factors for NHL through in-person interviews. Participants were asked to report jobs held for a year or longer during their lifetime. For each reported job, detailed information was elicited on job title, activities or duties, company name, type of business, year begun, and year ended. All participants reported their lifetime job history. On average, both cases and controls held 3–4 jobs during their lifetime. The average numbers of years worked were 27.4 for cases and 27.0 for controls. Jobs were coded according to the 1980 Standard Occupational Classification Manual (27) and the 1987 Standard Industry Classification Manual (28) scheme.

Exposure to organic solvents and formaldehyde associated with each job was assessed by linking the coded occupational data with a job-exposure matrix updated by industrial hygienists at the National Cancer Institute (29, 30). We assessed exposure to any organic solvent and to benzene, any chlorinated solvent, and 7 specific chlorinated solvents—dichloroethylene, chloroform, dichloroethane, dichloromethane, methyl chloride, trichloroethylene, and carbon tetrachloride—individually.

The job-exposure matrix was applied as follows: first, every occupation and industry was assigned a semiquantitative estimate of intensity and probability according to a scale of 0–3 (0=none, 1=low, 2=medium, and 3=high exposure). Intensity was estimated on the basis of expected exposure level and frequency of the specific substance for a worker in that industry or occupation. Exposure probability is the likelihood that the specific substance is used by a worker in a given industry or occupation. Second, occupational and industrial exposure scores were combined by using the following algorithms (31): If the exposure depended on occupation only, then the exposure score for the occupation-industry combination was the square of occupational exposure estimation, that is, intensity=intensity²_occupation, and probability=probability²_occupation. If the exposure was dependent on occupation and industry, then the exposure score for the occupation-industry combination was the product of the occupational exposure estimation and the industrial exposure estimation, that is, intensity=intensity_occupation × intensity_industry, and probability=probability_occupation × probability_industry. Third, for each subject, this information was combined with duration data to estimate 1) ever exposure status, that is, whether the exposure estimate was ever greater than 0 during the subject's lifetime; 2) cumulative intensity of exposure, defined as the sum of (intensity index for each job × correspondent job duration); 3) cumulative probability of exposure, that is, the sum of (probability index for each job × correspondent job duration); 4) average intensity of exposure, defined as the sum of (intensity index for each job × correspondent job duration)/exposure duration; and 5) average probability of exposure, that is, the sum of (probability index for each job × correspondent job duration)/exposure duration. The final scores for average exposure intensity and probability were categorized as never exposed (0), low (<3), medium (3–5), and high intensity/probability (≥6). Exposure within 1 year before diagnosis/interview was excluded from the exposure assessment. During the exposure assignment process, the disease status of individuals was masked.

Table 2 presents the association between solvent exposure and the risk of NHL by average intensity and by average probability. A significantly increased risk of NHL was observed for those with average probability of exposure to any organic solvent at a medium-high level (OR=1.5, 95% CI: 1.1, 1.9). Those exposed to chlorinated solvent with average intensity at a medium-high level had a significantly increased risk of NHL (OR=1.5, 95% CI: 1.1, 2.1). We observed borderline increased risks of NHL for those with exposure to any organic solvent at a medium-high average intensity level, to formaldehyde at a low average intensity or probability level, to chlorinated solvent at a low average probability level, and to dichloromethane at a low average probability level. The risk of NHL increased with the average probability or average intensity level of any organic solvent and with average probability level of chlorinated solvent. A 2.3-fold significantly increased risk of NHL was observed for those exposed to carbon tetrachloride at a low average intensity (95% CI: 1.2, 4.5) or a low average probability level (95% CI: 1.3, 4.0); in addition, the risk of NHL increased with average probability level of carbon tetrachloride. However, no participant was exposed to carbon tetrachloride on average at a medium-high probability level.

Table 3 illustrates the results adjusted by both average exposure intensity and probability. Those women exposed to any organic solvent at both medium-high average intensity and medium-high average probability levels had a significantly increased risk (OR=1.5, 95% CI: 1.1, 2.1). For those exposed to formaldehyde, a significantly elevated risk was observed at only a low average intensity and low average probability level (OR=1.4, 95% CI: 1.1, 1.9). Low average probability of exposure to chlorinated solvent accompanied by medium-high average intensity also significantly increased the risk of NHL (OR=1.7, 95% CI: 1.1, 2.4). A 2.3-fold increased risk of NHL was observed for those exposed to carbon tetrachloride at both a low average probability and a low average intensity level (95% CI: 1.2, 4.5).

We also evaluated associations by cumulative intensity and probability (data not shown). About 50% increased risks of NHL were observed for those whose cumulative exposure intensity to any organic solvent was at least 36 indexes × year (95% CI: 1.1, 2.0) and those whose cumulative exposure probability to organic solvent was at least 29 indexes × year (95% CI: 1.1, 2.0). An odds ratio of 1.6 was observed for those whose cumulative exposure intensity to chlorinated solvent was more than 35 indexes × year (95% CI: 1.1, 2.3). For specific chlorinated solvents, a significantly increased risk was observed for those exposed to carbon tetrachloride at a cumulative exposure intensity of 3–18 indexes × year (OR=4.3, 95% CI: 1.9, 10.2) and with a cumulative exposure probability of more than 35 indexes × year (OR=5.2, 95% CI: 2.0, 13.9). Moreover, the risk of NHL increased with increasing cumulative intensity exposure levels of organic solvent (P=0.01) and chlorinated solvents (P= 0.02). The risk of NHL also increased with cumulative exposure probability level of organic solvents (P < 0.01).

We further analyzed associations between occupational exposures and risks of major subtypes of NHL (Table 4). Analyses of risks of diffuse large B-cell lymphoma (DLBCL) showed the same patterns as the results for overall NHL. In addition, a significantly increased risk of DLBCL was observed for those exposed to benzene at an average medium-high probability level. Furthermore, risk of DLBCL increased with average intensity or probability level of benzene exposure. The effect of chlorinated solvent exposure on risk of NHL varied with major NHL subtypes at a high-medium average intensity level (χ²=7.24 (degrees of freedom=2), P=0.03), with a 2-fold increased risk of DLBCL (OR=2.3, 95% CI: 1.5, 3.6). The risk of DLBCL also increased with average intensity level of any chlorinated solvent (P < 0.01).

Our results also showed that the increased risk of NHL associated with exposure to solvents was seen mainly for those who reported exposure to chlorinated solvents. Chlorinated solvents have been widely used as extraction solvents, paint solvents, and coating solvents, as well as in cleaning and degreasing solvents. Seidler et al. (37) reported a significant association between malignant lymphoma risk and high exposure to chlorinated hydrocarbons (OR=2.1, 95% CI: 1.1, 4.3). McDuffie et al. (38) observed a 2.4-fold increased risk of NHL for male Canadian pesticide workers who may have been exposed to carbon tetrachloride (95% CI: 1.2, 5.1). The lymphatic system has been regarded as a target for trichloroethylene toxicity (39). Results of 2 cohort studies of aircraft maintenance workers exposed to trichloroethylene or perchloroethylene suggested an increased NHL risk (40, 41). Note that only a relative small number of participants in our study were ever exposed to carbon tetrachloride; moreover, this exposure occurred for the most part at a low average probability exposure level. Thus, the significant finding between carbon tetrachloride and risk of NHL in this study might be due to chance.

A dose-dependent association between benzene exposure and risk of NHL was observed. A significantly increased risk of DLBCL for those exposed to benzene at a medium-high average probability level was found. A nonsignificantly but increased risk of NHL was also found for those who had an average medium-high intensity or probability level of exposure to benzene. Lymphocytes in peripheral blood are a hematologic cell type particularly sensitive to benzene toxicity (42, 43), and peripheral lymphatic cells are targeted by the genotoxic metabolites of benzene (44). Furthermore, benzene exposure has been linked to chromosomal aberrations found in NHL in cultured peripheral lymphocytes of exposed workers (45).

Thus, our results are consistent with an association between benzene exposure and risk of NHL. However, results regarding the association between benzene exposure and risk of NHL have been inconsistent in previous studies. A Chinese cohort study reported a 4-fold increased risk of NHL (risk ratio=4.1, 95% CI: 1.2, 14.4) for workers employed before 1972, and the risk was strongly associated with cumulative exposure that occurred more than 10 years prior to diagnosis (P=0.005). A case-control study from France also reported an almost 5-fold increased risk of NHL (OR=4.6, 95% CI: 1.1, 19.2) for those who reported being exposed to benzene for more than 810 days (46). There are also studies that did not find any association between benzene exposure and risk of NHL among petroleum workers (47), which may have been due to the healthy-worker effect (48). As presented by Smith et al. (48), after adjusting for the healthy-worker effect, the observed standardized mortality ratio of lymphoreticulosarcoma in male refinery workers increased from 1.1 (95% CI: 0.6, 1.7) to 1.5 (95% CI: 1.2, 1.8). However, a recent meta-analysis found evidence of a significant association between occupational benzene exposure and NHL risk, particularly when restricted to studies with a higher-quality exposure assessment (49). Risk estimates for medium-high probability of exposure to benzene and NHL in the current study are consistent with the magnitude of effects reported in the meta-analysis.

Exposure to formaldehyde was found to be associated with an increased risk of NHL in our study, but the risk was mainly for those with a low exposure intensity or probability. Formaldehyde is used widely in manufacturing and chemical industries and also as a human tissue preservative. In experimental studies, formaldehyde is associated with increased frequencies of micronuclei, sister chromatid exchanges, chromosomal aberrations, and DNA-protein cross-links in peripheral lymphocytes of humans, as summarized by Hauptmann et al. (50). Two cohort studies have linked formaldehyde exposure to an increased risk of leukemia (50, 51). One epidemiologic study suggested that exposure to formaldehyde may be responsible for the observed increased risk of lymphoma for woodwork-related exposures (52); however, other studies have not found a positive association between formaldehyde exposure and NHL risk (13, 38, 53, 54).

Our results provide weak evidence that NHL risk may differ by subtypes. NHL comprises a heterogeneous group of lymphoid malignancies; it has been hypothesized that each subtype may have a distinct etiology (55). We observed that the effect of chlorinated solvent exposure on risk of NHL varied with major NHL subtypes at only a high-medium average intensity level. In this study, the sample size of a certain histologic subtype was relatively small, which limited our ability to detect the effect of solvent exposure on the risk of a certain NHL subtype. Few studies so far have evaluated the association between organic solvent exposures and risk of subtypes of NHL. Tatham et al. (53) found that exposure to solvents is associated with a significantly increased risk of small-cell diffuse lymphomas (OR=1.6, 95% CI: 1.1, 2.2). Rego et al. (20) reported a 2-fold increased risk of diffuse NHL (OR=2.0, 95% CI: 1.1, 3.7) for those whose first exposure to solvents occurred 6–25 years preceding diagnosis or interview. Fritschi et al. (36) reported an increased risk of B-cell NHL among workers ever exposed to solvents (OR=1.3, 95% CI: 1.0, 1.7). Kato et al. (56) found an increased risk of B-cell NHL (OR=1.5, 95% CI: 1.1, 2.1) for those with paint thinners/turpentine exposure.

This study is one of the few that has applied a job-exposure matrix to evaluate the association between solvents exposure and the risk of NHL in the United States. Most of the early epidemiologic studies used occupational titles or industry as surrogates to evaluate the association between occupational exposures and NHL risk. As pointed out by others, exposure and exposure intensity may vary by industries included in the same occupation title. Thus, use of job titles or industries separately as surrogates to study the exposure and disease relation may result in serious misclassification of exposure and intensity of exposure, which could bias the estimated association between the exposure and disease of interest (57). The advantages of using job-exposure matrixes to analyze occupational exposure and disease risk include that the job-exposure matrixes link information from both occupational titles and industry titles with specific exposures, therefore minimizing exposure misclassification, and allow for semiquantitative measurements of the occupational exposures and disease relation. Use of job-exposure matrixes in occupational epidemiologic studies could also significantly increase the statistical power compared with use of job or industrial titles (58).

Several potential limitations should be considered in interpreting our study results. One concern is misclassification of exposure. We assessed occupational exposure to specific solvents by linking the job-exposure matrixes with self-reported occupational histories, and some degree of exposure misclassification was unavoidable. In general, misclassification of exposure is unlikely to be differential. For multilevel exposure, nondifferential misclassification may bias relative risks away from the null in some categories (59, 60); however, it can bias the relative risk toward the null in the highest exposure category only (60). In this study, when exposure intensities and probabilities were categorized into more than 2 levels, significant associations were observed mainly at the highest exposure level.

Another potential limitation of the study is the relatively small sample size of certain subtypes and the low prevalence of occupational exposure to some organic solvents. A small sample size with low exposure prevalence not only reduces statistical power but also increases the possibility of false-positive associations. When we evaluated associations between occupational exposures and risks of major subtypes of NHL, significantly increased associations were observed for DLBCL patients only, the most common NHL subtypes in this study. Further studies with a larger sample size are needed to explore the associations between occupational solvents exposure and the risk of NHL by subtype.