Occupation and Risk of Non-hodgkin Lymphoma and Its Subtypes

A Pooled Analysis From the InterLymph Consortium

Andrea 't Mannetje; Anneclaire J. De Roos; Paolo Boffetta; Roel Vermeulen; Geza Benke; Lin Fritschi; Paul Brennan; Lenka Foretova; Marc Maynadié; Nikolaus Becker; Alexandra Nieters; Anthony Staines; Marcello Campagna; Brian Chiu; Jacqueline Clavel; Silvia de Sanjose; Patricia Hartge; Elizabeth A. Holly; Paige Bracci; Martha S. Linet; Alain Monnereau; Laurent Orsi; Mark P. Purdue; Nathaniel Rothman; Qing Lan; Eleanor Kane; Adele Seniori Costantini; Lucia Miligi; John J. Spinelli; Tongzhang Zheng; Pierluigi Cocco; Anne Kricker


Environ Health Perspect. 2016;124(4):396-405. 

In This Article


Study Population

Included in our analyses were 10 NHL case–control studies that participate in the InterLymph consortium, had collected information on occupation from cases and controls, and were willing to contribute their data to the pooled analysis (see Table 1 for the acronyms used to refer to each study, details about study designs and locations, and citations to general references for each study). The InterLymph consortium of international investigators undertakes research projects to pool data across studies that explore the etiology of lymphoid malignancies. The set of harmonized core variables, including age, sex, study center (region), smoking status, and NHL subtype, was directly obtained from the InterLymph data coordinating center. Variables on occupational history were obtained from the principal investigators of each participating study. We applied the lymphoma classification scheme for epidemiologic research developed by InterLymph investigators (Morton et al. 2007) to all participating InterLymph studies. All cases classified as "lymphoid neoplasms" according to this classification, except multiple myeloma and Hodgkin lymphoma, were included in this analysis.

Occupational History

For the purpose of our pooled analyses, the data on occupation were classified into a standard internationally recognized occupational classification scheme, the International Standard Classification of Occupations 1968 (ISCO-68) [International Labour Office (ILO) 1981]. Depending on the original occupational classification used by the individual studies and on whether a full-text description of the occupation was available, the ISCO-68 code for each job recorded was determined by one of the following methods: a) a direct conversion of the original classification to the ISCO-68 classification (for the Yale and UCSF1 studies); b) a direct conversion from the original classification to the ISCO-68 classification followed by checking the correctness of each ISCO-68 code by comparing it with the free-text information on the occupation (for the NCI-SEER study); c) using the free-text information on the occupation to individually assign the ISCO-68 code (for the BC, Nebraska, UK, and NSW studies); or d) directly using the original occupational codes for those studies that used ISCO-68 as their original classification (for the Epilymph, Italy, and ENGELA studies). Eight of the 10 studies collected the full occupational history of cases and controls including all occupations held for at least 1 year and starting and ending years, and 2 studies (Nebraska, BC) recorded only the longest-held occupation.

We defined occupational groups of a priori interest for NHL based on the peer-reviewed literature (Table 2). After discussions among three of the authors (A.'tM., A.J.D., R.V.), 25 occupational groups were constructed that included jobs associated with NHL in previous studies other than the 10 case–control studies included in our pooled analysis.

We also studied occupations within a group separately up to the detail of the 5-digit ISCO-68 code to explore whether an association was restricted to specific occupations within the group. For example, crop farmers were studied as a group, and specific occupations within this group such as orchard farmers and rice farmers were also studied separately.

Statistical Analyses

Unconditional logistic regression was used to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for the association between NHL and occupations in the pooled data set in models adjusted for age, sex, and study center. For each a priori occupational group and individual ISCO-68 occupation defined by a 1-digit, 2-digit, 3-digit, and 5-digit code, a dichotomous variable was created for ever having worked in that occupation. Duration of employment was coded as < 1 year, 1–10 years, and > 10 years in the occupation. Smoking status (never/former/current) was considered as a potential confounder, but adjusting for smoking made no substantial difference to the relative risk estimates (data not shown); consequently, smoking was not included as a covariate.

Analyses were performed for all NHL combined (excluding Hodgkin lymphoma and multiple myeloma) and separately for each of four major NHL subtypes [diffuse large B-cell lymphoma (DLBCL), follicular lymphoma (FL), chronic lymphocytic leukemia/small lymphocytic lymphoma (CLL/SLL), and peripheral T-cell lymphoma (PTCL)]; the same set of controls that was used for all NHL combined was used for each subtype. Two studies did not include CLL/SLL (UCSF1; UK) and were excluded from all CLL/SLL–specific analyses. All analyses were repeated stratified by sex. All statistical tests were two-sided with a significance level of 0.05. The Nebraska and BC studies included longest-held occupation only and were excluded from analyses of duration but were included in analyses of ever employment because their exclusion made little difference to the results.

Polytomous regression was used to test whether differences in ORs by NHL subtype were statistically significant at p < 0.05; we tested for heterogeneity in effect across the four subtypes (DLBCL, FL, CLL/SLL, PTCL) based on data for ever employment in the occupation with both sexes combined. We tested for heterogeneity among studies using Cochran's chi-squared test or the Q-test (Higgins and Thompson 2002); there was no evidence of significant heterogeneity (data not shown). To identify those associations with the largest potential impact on NHL incidence under the assumption of causality and in the absence of confounding, we calculated a population attributable fraction (AF) for occupations in which 1% or more of cases had ever worked and that were associated with an increased relative risk. The formula for AF calculation used the prevalence of ever employment in each occupation in controls as an estimate of population prevalence: prevalencecontrols × (OR – 1)/[1 + prevalencecontrols(OR – 1)] (Last et al. 1995).

Criteria for Presentation of Results

The present analysis involved many specific occupations within the 25 a priori groups for which previous research demonstrated an association with an increased relative risk of NHL: 925 of > 2,000 relevant codes in the ISCO-68 classification were involved in this analysis. We set criteria to determine which associations to include in the results. We present results for ever employment and for > 10 years employment for all NHL and each of the four subtypes for each occupational group of a priori interest regardless of whether the estimates were statistically significant, with the exception of occupational groups with < 10 cases or < 10 controls. One occupational group in the analyses of all NHL (undertakers) and two groups in the analyses of the four subtypes (pulp & paper workers and petroleum workers) were excluded from the results because they had < 10 cases or < 10 controls. Additionally, we report associations with specific occupational titles included within the occupational groups of interest if we estimated a statistically significant OR (> 1.10 or < 0.90, for ever employment or > 10 years employment) based on men and women combined for all NHL or for any one of the four subtypes.

ORs were also calculated for the 1,286 occupations that were not included in the 25 groups of a priori interest. These results are not presented here but are available upon request.