Non-Hodgkin’s lymphoma (NHL) represents a diverse group of immunoproliferative diseases, the majority with a B lymphocyte origin. NHL is the sixth most common cancer in New Zealand. Incidence has been increasing steadily since the 1950s,1 with some indication of a levelling off in recent years, a trend also observed in other developed countries.2

The aetiology of NHL and the causes behind its increasing incidence are largely unknown. Immune deficiency, for example as seen in people with AIDS and transplant recipients, is a known risk factor for NHL.3 Some viral infections, especially by Epstein–Barr virus—a herpes virus with B cell-transforming activity—have, in addition, been associated with an increased risk of NHL.3 Epidemiological studies of occupational risk factors for NHL have suggested associations between several chemical exposures and NHL, with the evidence being most consistent for pesticides and chlorinated solvents. It has therefore been suggested that occupational exposures may have contributed in part to the increasing incidence.4

In New Zealand, the only occupational studies of NHL date from the 1980s. These studies showed increased risks for farmers,5^–7 meat workers8 9 and forestry workers.10 Here we present results from a new nationwide case–control study of 291 NHL cases diagnosed in New Zealand during 2003 and 2004. The main aims of the study were to assess whether previously reported associations persist, and to identify other occupations that may also contribute to the risk of NHL in the New Zealand population.

METHODS

This study is part of a series of three linked case–controls studies of NHL, leukaemia and bladder cancer, all of which have used the same group of population controls. The findings for leukaemia and bladder cancer will be reported elsewhere.

Potential participants in the study were all incident cases of NHL, aged 25–70 years, notified to the New Zealand Cancer Registry during 2003 and 2004, a total of 553 notifications nationwide. Both the treating clinician and general practitioner (GP) of the patient were sent a letter explaining the study and asking for consent to contact the patient. For 15.9% of the notifications, either the clinician or the GP did not provide consent to contact the patient. Of the 464 remaining cases, for 89 no contact could be established by mail and a further 40 were not eligible (eg, never worked in New Zealand, mental health problems, NHL is not primary cancer). From the 335 remaining cases, 44 (13%) declined to participate and 291 cases were interviewed for the study. Eight of these were next of kin interviews. Thus, if those known to be ineligible for the study are excluded, the response rate was approximately 69%.

Controls were randomly selected from the New Zealand Electoral Roll for 2003, frequency matched by age according to the age distribution of New Zealand cancer registrations for NHL, bladder cancer and leukaemia in 1999. A letter of invitation was sent to 1200 individuals, of which 100 were returned to sender and thus considered ineligible. Of the remaining 1100, for 348 (32%) contact could not be established. Their addresses were subsequently compared with the most recent Electoral Rolls of 2005 and 2006. Of the 348 non-responders, 20 did not appear or appeared with another address on the new Electoral Roll and were thus considered ineligible. Of the 752 for whom contact could be established, 92 were ineligible because of other reasons (eg, never worked in New Zealand). Of the remaining 660 controls, 187 declined to participate (28%), and 473 population controls were interviewed. Thus, if those known to be ineligible for the study are excluded, the response rate in the controls was approximately 48%.

The interview was conducted face-to-face at the home of the case or control with a trained interviewer with an occupational health nursing background. The questionnaire collected information on demographics, smoking and a full occupational history. Each job held since leaving school was listed, including the start year, year of termination, department and job, and name, location and activity of the employer. Then, for each job with a minimum duration of 12 months, more detailed questions were asked, including a task description, use of machines and materials, self-reported exposures, workplace ventilation and use of protective equipment.

For the purpose of the analyses presented here, each job was coded according to the 1999 New Zealand Standard Classification of Occupations (NZSCO 1999)11 (hereafter referred to as the occupational code) and the Australian and New Zealand Standard Industrial Classification (New Zealand use version 1996)12 (hereafter referred to as the industry code). The occupational code was based on the full job and task description, rather than on the occupational title alone, to ensure that the code covered the actual tasks of each job. The industry code was based on the activity of the employer. All coding was done blind to the case–control status of the participants.

Before the data analyses were conducted, a broad list of a priori high risk occupations was constructed, based on the international literature, which included farmers, meat workers, painters, metal workers and welders, machinery mechanics and operators, drivers, printers, textile workers, leather and shoe workers, health professionals, teaching professionals, hairdressers, fire fighters, wood workers, funeral directors, and bakers and grain millers.

Unconditional regression using SAS V9.1 was applied to estimate the odds ratio (OR) and its 95% CI for ever being employed in a certain occupation/industry, compared with never being employed in that occupation/industry. ORs were calculated for all 958 occupational codes. Of these 958 codes, only 235 had 10 study subjects or more that ever worked in these occupations, and only results of these 235 occupations were evaluated. ORs were also calculated for each industry code. Of these 684 codes, only 228 contained 10 subjects or more.

ORs were adjusted for age (5 year age groups), Māori ethnicity, sex and smoking (never, ex, current). Cases and controls were considered current smokers if they reported to have stopped smoking less than 2 years before the interview. Logistic regression models were also adjusted for occupational status, based on the New Zealand Socioeconomic Index of Occupational Status (NZSEI)13 (continuous variable ranging between 20 and 90) of the longest held occupation. Whether a longer duration in a certain occupation was associated with an increased risk was studied through categorical variables for duration of each job (1–2 years, >2 to 10 years, >10 years). A test for trend for duration was performed by fitting this categorical variable as a continuous variable in the model.

Semi-Bayes adjustment

Because of the large number of occupations and industries being considered, this type of study carries the risk that some of the findings involving elevated ORs will be due to chance. A semi-Bayes (SB) approach was therefore applied to determine which of the findings were the most robust.14 The basic idea of empirical Bayes (EB) and SB adjustments for multiple associations is that the observed variation of the estimated relative risks around their geometric mean is larger than the variation of the true (but unknown) relative risks. In SB adjustments, an a priori value for the extra variation is chosen so that the true relative risks have a reasonable range of variation, and is then used to adjust the observed relative risks.15 The adjustment consists of shrinking outlying relative risks towards the overall mean (of the relative risks of all the different “exposures” being considered). The larger the individual variance of the relative risks, the stronger is the shrinkage—that is, the shrinkage is stronger for less reliable estimates based on small numbers. Typical applications in which SB adjustments are a useful addition to traditional methods of adjustment for multiple comparisons are large occupational surveillance studies, where many relative risks are estimated with few or no a priori beliefs about which associations might be causal.15 SB estimates were calculated using R (free software for statistical computing and graphics).16 The input for SB adjustments were the maximum likelihood estimate of β (logOR), resulting from the multivariate logistic regression for each occupation and industry. The variance of the true logOR was assumed equal to 0.25. Assuming a normal distribution of the logORs, this choice implies that the true ORs are within a sevenfold range of each other.14

For those occupations (or industries) which were not considered to be of a priori high risk for NHL, estimates were shrunk towards the mean for all occupations (or industries). Similarly, for those occupations (or industries) which were considered to be of a priori high risk for NHL, estimates were shrunk towards the mean for all such occupations (or industries).

The findings for all occupations and industries, both before and after SB adjustment, are available on web-based tables (see Supplementary data).17 Here we report the findings for a priori high risk occupations and industries, and for other occupations and industries that showed statistically significant elevated or decreased risks in the analyses.

RESULTS

The study included 291 interviews with NHL cases, and 473 interviews with population controls. Of these, two controls were excluded due to missing values in key variables, leaving 291 cases and 471 controls available for analysis (table 1). Cases were 54% male (46% female) and controls were 47% male (53% female), with a mean age of 56.8 in cases and 59.2 in controls. Eleven cases and 14 controls reported Maori ethnic background. Current smoking was more common in the cases (16%) than in the controls (8%). Occupational class distribution was similar for cases and controls, except for the lowest occupational class (class 6), which had a higher frequency in the cases (36%) than in the controls (24%).

We studied whether this difference in occupational class between cases and controls could have been a result of response bias in the controls—that is, that controls with lower occupational class were less likely to participate in the study. For this purpose we compared the sex, age and occupational class distributions between the 471 participating controls and the 729 non-participating controls using the information available from the Electoral Roll. This showed that both sex and age were significant determinants of non-participation within the controls, with men and younger ages less likely to participate. Logistic regression showed that the lowest occupational class (class 6) was a statistically significant determinant of non-participation in controls (OR 1.81, 95% CI 1.17 to 2.81), adjusting for age and sex, while all other occupational classes had ORs for non-participation of 1.02–1.10 compared with the highest occupational class. Logistic regression models were therefore also adjusted for occupational class.

A priori high risk occupations and industries

Tables 2 and 3 list the findings for the a priori high risk occupations and industries, respectively, both adjusted for and stratified by sex (but not SB adjusted).

Farming and agriculture

Employment as an agricultural or fishery worker (table 2 occupational code 61) was not associated with an increased risk for NHL (OR 1.04, 95% CI 0.73 to 1.47). Within the specific farming occupations, an increased risk for NHL was observed for field crop and vegetable growers (OR 2.74, 95% CI 1.04 to 7.25) and nursery growers (OR 3.16, 95% CI 1.03 to 9.69). An increased NHL risk was not observed for the subgroup of livestock producers (OR 0.65, 95% CI 0.36 to 1.16).

The results for employment in the farming industry (table 3) were similar to those observed for farming occupations, with statistically significant increased risks for horticulture and fruit growing (OR 2.28, 95% CI 1.37 to 3.79), particularly for plant nurseries (OR 4.30, 95% CI 1.08 to 17.2), vegetable growing (OR 2.32, 95% CI 0.90 to 6.00) and apple and pear growing (OR 4.91, 95% CI 1.26 to 19.1). A reduced NHL risk was observed for grain, sheep and beef cattle farming (OR 0.56, 95% CI 0.29 to 1.08) and dairy cattle farming (OR 0.55, 95% CI 0.29 to 1.07). However, other livestock farming (industry code A015) was associated with an increased NHL risk (OR 9.75, 95% CI 2.04 to 46.5) for men and women. Comparing the ORs for the whole agriculture industry (industry code A01) between men and women revealed a statistically significant increased risk for women (OR 1.72, 95% CI 1.01 to 2.92), while no increased risk was observed for men in the agricultural industry (OR 0.88, 95% CI 0.53 to 1.46).

Analyses by duration of employment in the large occupational group of farmers and farm workers (market oriented agricultural and fishery workers) showed a statistically significant increased risk for the relatively short-term workers (1–2 years: OR 2.64, 95% CI 1.27 to 5.48), while no increased risk was observed for longer term farmers. No consistent pattern of duration–response was observed for any of the specific farming occupations, but numbers were small.

Meat workers

In total, 6.9% of the cases and 3.4% of controls had ever worked as a slaughterer, which was associated with an increased risk for NHL (OR 1.81, 95% CI 0.97 to 3.97). Risk was increased for all three duration strata without a clear duration–response pattern. Risk was increased for both men and women, and the same pattern was observed for the meat processing industry (table 3).

Painters

An increased risk of NHL was observed for painters, decorators and paperhangers (OR 2.45, 95% CI 0.87 to 6.85) (table 2), without a clear association with duration of employment.

Metal workers

Statistically significant increased risks were not observed for any of the metal working, welding occupations or machinery mechanics and operators (table 2), although ORs were elevated for most of these a priori high risk occupations, particularly for male metal-processing plant operators (OR 2.64, 95% CI 0.84 to 8.31), who process metal by heating, casting, rolling, drawing and extruding, and treat metal with chemicals. Results by industry did show a statistically significant increased risk for NHL in association with metal product manufacturing (OR 1.92, 95% CI 1.12 to 3.28) (table 3).

Machinery mechanics and operators

Motor mechanics did not have an elevated NHL risk (OR 0.76, 95% CI 0.34 to 1.70). No increased risk was observed for machinery and equipment manufacturing (table 3), but those working more than 10 years in this industry did have a statistically significant increased risk for NHL (OR 3.49, 95% CI 1.16 to 10.5).

Drivers

Motor vehicle drivers in general did not have an increased risk for NHL (OR 0.92, 95% CI 0.51 to 1.66) (table 2), but heavy truck drivers did show an increased risk (OR 1.98, 95% CI 0.92 to 4.24), being statistically significant for male heavy truck drivers (OR 2.44, 95% CI 1.10 to 5.41). The same pattern was observed for the transport industry (table 3), which showed an increased risk for road freight transport (OR 2.26, 95% CI 0.90 to 5.67) but not for road passenger transport (OR 0.74, 95% CI 0.29 to 1.92).

Other a priori high risk occupations

The a priori high risk occupations and industries of printers, textile workers, healthcare workers, teachers, hairdressers and wood workers did not show increased risks of NHL, in either men or women (tables 2 and 3). Risks by duration did not show a clear pattern for any of these groups, except for textile, clothing, footwear and leather manufacture (industry code C22) for which employment of longer than 10 years was associated with a statistically significant increased risk (>10 years 11ca/6co OR 3.64, 95% CI 1.26 to 10.5).

SB adjustment of the a priori high risk occupations or industries

Ever being employed in one or more of the a priori high risk occupations (table 2) and industries (table 3) was associated with only a slight increased risk for NHL (OR_{a priori occupation} 1.13, 95% CI 0.81 to 1.56; OR_{a priori industry} 1.09, 95% CI 0.76 to 1.57). All estimates in tables 2 and 3 were also regressed towards this mean using SB adjustment. This generally resulted in an attenuation of the ORs, and none of the ORs for the a priori high risk occupations remained statistically significant at the p<0.05 level after SB adjustment. Two industry sectors remained statistically significant (p<0.05) after SB adjustment, namely horticulture and fruit growing (OR_SB 1.94, 95% CI 1.21 to 3.11) and metal product manufacturing (OR_SB 1.68, 95% CI 1.03 to 2.74).

SB adjustment of the a posteriori high risk occupations or industries

Occupations and industries with an observed increased or decreased risk (p<0.05), but not considered an a priori high risk occupation, are listed in table 4. Client information clerks formed the only occupation with a statistically significant decreased risk for NHL (OR 0.41, 95% CI 0.21 to 0.80), which remained after SB adjustment (OR_SB 0.55, 95% CI 0.31 to 0.98). Four occupations showed a statistically significant increased risk (see table 4), in addition to the a priori high risk occupations listed in table 2. Two of these remained statistically significant after SB adjustment: the general occupational group of labourers and related elementary service workers (OR_SB 1.64, 95% CI 1.15 to 2.34) and its subgroup of cleaners (OR_SB 1.80, 95% CI 1.11 to 2.94).

A statistically significant decreased risk was observed for three industries and an increased risk for two industries (see table 4). After SB adjustment, none of these ORs remained statistically significant. Results by sex indicated an additional a posteriori high risk industry for men: construction trade service (industry code E42) (OR_SB 1.72, 95% CI 1.05 to 2.83). This industry includes bricklaying, plumbing and painting services.

DISCUSSION

This study of 291 incident NHL cases diagnosed in New Zealand during 2003 and 2004 and 471 population controls aimed to identify occupations that entail an elevated risk for NHL in New Zealand. After adjustment for age, smoking status, Māori ethnicity and occupational status, this study showed that farmers, particularly those in field crop, vegetable, horticulture and fruit growing, and meat workers remain at high risk for NHL, and that several other occupations and industries (including metal product manufacturing industry, painters, cleaners, heavy truck drivers) also have an increased risk for NHL in the New Zealand population. SB adjustment indicated the most robust findings of this study; an increased risk of NHL for horticulture and fruit growing, metal product manufacturing and cleaners.

The use of the Electoral Roll as a source of population controls for this study held certain advantages and disadvantages. All New Zealand citizens and permanent residents aged 18 years and older are legally required to enrol on the New Zealand Electoral Roll. Although using the most up to date Electoral Roll available at the time the study started (2003), this still resulted in a high percentage of selected controls that could not be contacted either by mail or by phone. The use of the Electoral Roll to select controls has, however, the advantage that occupation is also available for non-participating controls. Comparison of participating and non-participating controls showed that participants were less often of the lowest occupational class, which led to the decision to adjust all associations for occupational class. This adjustment generally led to only slightly attenuated ORs, but did not alter the main results of this study.

Another disadvantage of any occupational study where multiple comparisons are made for many occupations is the risk that some findings may be elevated and/or statistically significant by chance. For this reason we calculated SB-adjusted estimates. This generally resulted in attenuation of the risk estimates towards the null, particularly for those estimates based on small numbers. Several risk estimates, however, remained statistically significant, which helped us to identify the most robust findings of our study.

The size of this study (291 cases, 471 controls) prohibited the evaluation of risk by NHL subtype, which is unfortunate as larger studies have shown that the aetiologies of different subtypes are likely to differ. For several of the occupations of a priori interest for NHL (fire fighters, wood workers, funeral directors, bakers and grain millers, agricultural sprayers, leather and shoe workers), the numbers were very small, and this study could not provide information as to whether they entail an increased NHL risk in New Zealand. However, for several common occupational groups that are of particular importance within the New Zealand work force (eg, agriculture and meat working) this study did provide sufficient power to produce valuable information.

Farmers

Previous New Zealand studies conducted in the 1980s showed an increased NHL risk for farmers,5 7 although these studies were limited to men only, and data on potential confounders and specific occupation exposures were limited. Many studies conducted in other countries have since found statistically significant positive associations for farmers, although several others have not.18 Pesticide exposure is one of the hypotheses proposed to explain increased NHL risk in farmers, but farmers can also be exposed to numerous other substances that may entail an increased risk for NHL, namely infectious agents from animals, diesel fumes, solvents and dusts.19

Our study showed a clear difference in NHL risk between crop growers on one hand and animal producers on the other. Employment as a market farmer and crop grower showed an increased risk for NHL, while we did not observe an increased risk for the main animal-producing sectors in New Zealand (sheep and dairy farming), although an increased NHL risk was observed for other livestock farming, representing a miscellaneous group of other farm animals including pigs, horses, deer and mixed livestock. These findings (increased risk for crop farmers and not for livestock farmers) are the opposite of what has generally been found overseas. A recent meta-analysis reviewing published NHL studies18 observed no increase in NHL risk for crop farming (RR_meta 0.96, 95% CI 0.83 to 1.09) while an increased risk was reported for livestock farming (RR_meta 1.31, 95% CI 1.08 to 1.60). Both meta-estimates were, however, strongly heterogeneous, which could be reflecting the diverse occupational circumstances of farmers in different countries. Our findings for farming may well be particular to New Zealand, given that very similar results were observed in a national study two decades previously.6 In that study, the largest relative risks for specific farming types were reported for orchard workers (OR 3.7, 90% CI 1.1 to 12.1), and no elevated risks were observed for sheep (OR 1.1, 90% CI 0.7 to 1.6) or dairy (OR 0.9, 90% CI 0.6 to 1.3) farming. New Zealand is likely to differ from other countries not only in terms of the main employing production sectors but also in terms of pesticide use and production methods. Crops farmed in New Zealand are predominantly perishable crops such as fruit and vegetables in which pesticides, especially insecticides and fungicides, play an essential part. The main livestock farming sectors in New Zealand are pasture land farming (dairy and sheep), which is typically non-intensive and therefore dissimilar to animal rearing in more densely populated countries. The group of “other livestock farming” for which we observed an increased NHL risk may be more similar to intensive animal rearing and be associated with a different set of exposures. The next step in our study will therefore involve exposure-specific analyses, aiming to clarify the marked difference in risk we observed between crop and animal farmers.

In our study, NHL risk was particularly elevated for female farmers and farm workers and less so for men. Again, this is somewhat different from what was found in other countries. A meta-analysis from 199820 reported a meta-estimate of 0.94 (95% CI 0.82 to 1.06) for female farmers based on 11 studies, none of which reported a significant elevation in relative risk. Some studies have nonetheless observed a difference in relative risk between men and women similar to our findings,21^–24 although the observed difference in our study may be due to chance. However, differences in exposure patterns between men and women may have contributed and will be studied in subsequent analyses.

Meat workers

This study observed a twofold increased NHL risk among meat workers, which is consistent with earlier studies in New Zealand.6 8 25 An increased risk of NHL has been observed for meat workers in other countries,26^–29 but not consistently.18 30 In New Zealand, the meat industry has a high production volume of mainly cows and sheep, and workers have physically demanding manual jobs that may entail high exposures to blood, faeces, urine and other biological exposures, as well as disinfectants. In a cohort study of 6647 New Zealand meat workers25 there were significant trends of increasing risk of lymphohaematopoietic cancer with increasing duration of exposure to biological material. The same study also showed an increased risk for lung cancer, and it was hypothesised that viral infections may be responsible for the increased risk for both cancer types in these workers, but a specific causal exposure has not yet been identified. Since our study showed no increase in NHL risk for sheep and dairy farming, our results imply a role for exposures specifically associated with the slaughtering, rather than exposures associated with the rearing of these animals.

Painters

Our study showed a more than twofold NHL risk for painters, a group representing mainly house painters. Case–control studies from Canada31 Sweden32 33 and the USA,34 35 and a Swedish cohort of male paint industry workers36 also reported a positive association for NHL and painters. In New Zealand, the majority of houses have a wooden exterior that requires regular painting, for which organic solvent-based paints were commonly used until recently. Exposure to organic solvents is thus a plausible explanation for this repeated finding, although painters can be exposed to other agents including paint and wood dust, asbestos, lead and wood preservatives.

Metal workers

In this study, workers in metal product manufacturing had a statistically significant increased risk for NHL. Increased NHL risk for occupations in metals and metal products has also been reported in several other studies.28 31 35 37 A variety of exposures could be present in the work environment, including metal dusts, metal fumes, metal working fluids (containing pesticides to prevent bacterial growth), asbestos, solvents and polycyclic aromatic hydrocarbons. Findings on the association between metal exposure and NHL are inconsistent,38 and exposure to solvents remains another plausible explanation for the observed increased NHL risk in metal product manufacturing.

Drivers

Our study showed an increased risk for heavy truck drivers and for the road freight transport industry. Interestingly, other drivers and the road passenger transport industry did not show increased risks. Truck driving has been associated with NHL in two Swedish studies.39 40 In another Swedish study,33 gasoline and oil products were associated with NHL, while diesel engine exhaust and gasoline engine exhaust were not. Studies among farmers showed positive associations with NHL for diesel fuel/exhaust41 and diesel fuel expenditure.42

Teachers

Two meta-analyses18 43 showed statistically significant increased but heterogeneous meta-estimates for teachers and NHL. It has been proposed that viral infections from the children could underlie this association,43 but this hypothesis has not been tested. Our study shows no increased risk of NHL for teachers, although a non-significant increased risk was observed for female primary and early childhood teachers (OR 1.69, 95% CI 0.82 to 3.47).

Other occupations and industries

In addition to considering a priori high risk groups, we also assessed the NHL risks in all other occupations and industries (table 4). Cleaners had an increased risk of NHL in our study which remained statistically significant after SB adjustment. We did not consider cleaners as an a priori high risk occupation for NHL, but increased risks have been observed for cleaners in an Italian44 and US28 study. Cleaners can be exposed to a variety of agents, including dusts, cleaning products, solvents, infectious agents and disinfectants.

The only occupation for which we observed a statistically significant decreased risk for NHL was client information clerks. This association remained statistically significant after SB adjustment, making it less likely to be due to chance. The finding is, however, not supported by other studies, and a causal exposure is not readily hypothesised. Contact with the public has in fact been studied as a risk factor for NHL, although an association has not been detected.45

In conclusion

This study observed a diverse list of high risk occupations for NHL largely in concordance with previous studies in New Zealand and elsewhere. Most notably, NHL risk was increased for the field crop, vegetable, horticulture and fruit growing industry, meat workers, painters, metal product manufacturing, truck drivers and cleaners. In this population, 23% of the controls were employed in one or more of these high risk occupations and industries, indicating that a large proportion of the New Zealand working population has been, and continues to be, employed in an occupation potentially entailing an increased risk for NHL.