Click on the links below to access all the ArticlePlus for this article.
Please note that ArticlePlus files may launch a viewer application outside of your web browser.
In its Monographs on the Evaluation of the Carcinogenic Risks in Humans, the International Agency for Research on Cancer has classified 29 occupational agents and 12 exposure conditions as carcinogenic.1,2 Several of these are lung carcinogens, including such widespread exposures as asbestos, polycyclic aromatic hydrocarbons (PAHs), silica, heavy metals, and environmental tobacco smoke (ETS). Since the end of the 1970s there has been a wide debate about the burden of lung tumors attributable to occupational exposures, and several estimates have been proposed.3–5 Few large-scale studies on this topic have been recently conducted in Europe.
We took advantage of the database of the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort to assess the risk of lung cancer associated with a history of employment in 1 or more selected occupations. Given the age range of the study participants, most of exposures presumably occurred during the 1970s and 1980s.
The EPIC is a multicenter European study, coordinated by the International Agency for Research on Cancer (Lyon), in which more than 500,000 healthy volunteers have been recruited in 10 European countries (Sweden, Denmark, Norway, Netherlands, the United Kingdom, France, Germany, Spain, Italy, and Greece).6 The cohort includes subjects of both sexes, in the age range of 35–74 years (median 53.3) at recruitment. Recruitment took place in 1992–1998. Dietary information on the frequency of consumption of more than 120 foods and drinks was obtained by a self-administered food frequency questionnaire, validated in a pilot phase. At enrollment, weight, height, and waist and hip circumferences were measured for each participant. Detailed information has been collected on reproductive history, physical activity, smoking and alcohol drinking history, medical history, education level, and other socioeconomic variables; the lifestyle questionnaire was printed in 2 separate versions for men and women. Both diet and lifestyle questionnaires were processed in the centers and a computerized central database has been developed after checking, coding, and quality control procedures. Present exposure to ETS was also investigated by questions on exposure to ETS (yes/no); place of exposure (home, work); and exposure to ETS during childhood.6 Only 11 of the 22 EPIC centers (in France, Italy, Denmark, Sweden, the Netherlands, and Potsdam, Germany) included questions on ETS in the questionnaire, and most of these collected information only on 1 or both of the first 2 items above.7 In Spain, Greece, Denmark, Germany, 4 Italian centers (Turin, Varese, Florence, Ragusa), and in 1 UK center (Cambridge) the lifestyle questionnaire included several questions on the occupational history of the participants, focusing on 52 selected jobs that have been previously linked to the risk of developing cancer. Many subjects reported having been employed in more than 1 at-risk occupation, and so the number of jobs related to different exposures held by each subject was also investigated.
The cohort has been followed-up since inception through cancer registries, vital statistics (mortality), active follow-up (France and Germany), and, in some areas, hospital discharge data. In 98% of the cases, diagnosis is based on histological confirmation.
The association between jobs and lung cancer incidence was analyzed by the Cox proportional hazard regression model, with age as the time variable and length of follow-up as a covariate. Due to potential confounding by several demographic and behavioral factors associated with both occupation and disease, hazard ratios (HRs) were stratified by country and adjusted for sex, smoking history (see below), highest school level reached (as a social class indicator), baseline values of fruit and vegetable consumption (linear trend), body mass index [weight (kg)/height (m2); linear trend], and leisure-time physical activity (active vs. nonactive). As cigarette smoking could exert a confounding effect on the risk estimates associated with all the occupations, we carefully controlled for lifelong smoking history. Instead of using a single variable indicating the cumulative dose (pack-years), we chose to control separately for age at initiation, mean lifelong intensity, and duration.8 Never-smokers were assigned a value of 0 for duration and intensity, and a value equal to the age at recruitment for the age at initiation. In addition, an indicator of current/former smoking at enrollment was included in the model. Subjects whose smoking status was unknown were excluded from the analysis (n = 5541). As expected, all covariates included in the model were associated with past employment in at least 1 of the 52 occupations.
Single jobs were analyzed separately, and individual HRs were computed adjusting for the above-mentioned covariates. The problem of inflation of type I error, due to the large number of occupations tested, was dealt with in 2 ways. The first consisted in computing, for each statistically significant HR, the false-positive report probability as described by Wacholder et al.9 This method is based on a Bayesian approach and has been originally applied mainly in the analysis of genetic polymorphisms. To obtain this probability, the investigator must assign to each exposure the “prior probability” that it is truly associated with the disease of interest. This “prior probability” is at least in part arbitrary, although it is based on knowledge deriving from earlier epidemiologic studies and biological plausibility. In the case of occupational exposures, the existing knowledge (both biological and epidemiologic) is vast and allows the use of generally high prior probabilities (in the order of 0.25 or more). A false positive report probability value below 0.2 is usually considered satisfactory.9
The second approach to minimize type I error was to reduce the number of statistical tests by combining different occupations into a limited number of exposure “scores.” First we created a global score by combining all the 52 selected occupations. This score was computed as the total number of occupations reported (as a continuous variable or categorized in 5 classes), or as yes/no (at least 1 occupation reported). Furthermore, we combined subsets of the 52 occupations according to presumed common carcinogenic exposures. Several job-exposures matrices, (specific for different countries) have been devised to estimate an individual's exposure to specific carcinogens given its occupational history.10,11 However, we judged that data included in our database were too nonspecific (lacking information about duration of employments and specificity of jobs) to fully exploit existing job-exposure matrices. Therefore we decided to obtain a consensus among a group of experts about a categorization of occupations according to a limited number of shared exposures, based on the Italian CAREX-JEM.12 We focused on the following carcinogenic exposures: asbestos (present in asbestos production, in construction and demolition industries, in shipyards, in insulation production, and in car repair stations); heavy metals (present in foundries, in metal industries, and in occupations related to welding, turning, and electroplating); PAHs (associated with refineries, asphalt, transportations, and car repair stations); ETS (particularly elevated for workers in bars and restaurants); silica (mostly present in mines, glass, and ceramic industries). Appendix 1 (available with the online version of this paper) shows how the jobs were combined in the specific scores. As for the global score, specific scores were computed as the sum of the reported jobs included in each group. Alternatively, the scores were treated as dichotomous (yes/no) or categorized in 3 or 4 classes. Finally, to assess the impact of the occupational exposures on the total lung cancer incidence, we computed the percent population attributable risk (PAR%) according to Breslow and Day.13 However, these figures should be considered with some caution, because the cohort may not be a representative sample of the general population in terms of prevalence of exposed subjects.
Baseline Characteristics of Cases and Noncases
The median follow-up for the 217,055 subjects enrolled was 6.1 years (interquartile interval, 4.9–7.0) and 809 lung cancer incident cases were recorded. The baseline characteristics for cases and for the whole cohort are reported in Table 1.
Single Occupation Analysis
As reported in Appendix 2 (available with the online version of the paper) the data for several jobs (nuclear industry, ceramics and dye production, etc.) were too sparse to have sufficient statistical power for interpretation. Eighteen occupations (Table 2) were associated with a statistically significant risk increase when men and women were analyzed together (assuming no association between any of the occupations and lung cancer; 2.6 significant results would be expected due to chance only). When the analysis was focused on men, 10 of the 52 occupations showed a statistically significant risk increase. Among women, some hazardous occupations (such as working in asbestos-cement production, in demolitions, constructions, and metal industry) were not considered due to the small numbers of women employed. Some new associations emerged in women, namely an increased risk related to agriculture (before 1960) and farming. The analysis of the false-positive report probability (Table 2) shows that, assuming a plausible HR of 1.5 and a prior probability of 0.5, all the positive results obtained in the cohort except 1 (working in leather and tanning industry) yield values below 0.2, and most do so even assuming a prior probability of 0.25. Conversely, the increased HR for agriculture workers before 1960 and for bus/taxi drivers, found in women only, should be considered with particular caution, given the low a priori probability and the high false-positive report probability.
Among single occupations, we specifically examined employment as a butcher or in slaughterhouses because the epidemiologic results reported in the literature are conflicting, and the underlying causal agent is still to be identified. We found increased risk for subjects employed as butchers, particularly women. As reported in Table 2 and Appendix 2, the HR for lung cancer in butchers was 1.75 considering both sexes together, and 2.2 in women. As potential confounding by smoking has been suggested for this category, we carefully controlled for smoking both by multivariable analysis and by stratification. A high risk for lung cancer did not seem to be limited to current smokers (adjusted HR = 1.60 [95% CI = 1.0–2.6] in current smokers; 1.59 [0.56–4.5] in former smokers; and 2.62 [0.4–20] in never-smokers).
Occupations Combined by Inferred Exposure
Table 3 shows the lung cancer HRs associated with scores as described above. All the HRs were stratified by country and adjusted for age, sex, educational level, smoking status, body mass index, and fruits and vegetable consumption. When computed as yes/no, the global score was associated with a HR of about 1.4. The test for heterogeneity of HRs across the different countries was nonsignificant (χ2 = 1.95, P = 0.86). A nonsignificant interaction was also found between smoking and the global score (P = 0.68) with HRs ranging from 1.31 and 1.51 in the 3 highest quartiles of pack-years and a slightly higher HR (2.17) in the lower quartile.
Except for silica, the HRs for the exposure-specific scores were larger than for the global scores, ranging from 1.4 to 1.6. Linear trends were highly statistically significant, with the exception of silica and PAH, with a clearly increasing risk with the number of at-risk jobs in most cases. After stratification for sex, elevated HRs were observed among women (even higher than in men) for the global score and for occupational ETS exposure.
As expected, we found a strong association between the occupational ETS score and self-reported exposure to passive smoking in the questionnaire: the odds ratio for an occupational ETS score of 1.0 or more when reporting passive smoke was 1.73 (95% CI = 1.61–1.86). Data on self-reported ETS exposure have been described elsewhere.7 Briefly, in a case-control study within the cohort study, an OR of 1.70 (95% CI = 1.02–2.82) was found for respiratory diseases in nonsmokers exposed to ETS at home or work. In the whole cohort, we found a HR of 1.65 (95% CI = 1.04–2.63) for lung cancer in nonsmokers exposed to ETS at work. Furthermore, daily exposure for many hours to ETS during childhood was associated with lung cancer in adulthood (HR = 3.6, 95% CI = 1.2–11.1).
The estimated attributable risk for the at-risk occupations, combined in scores, are reported in Table 4, separately for men and women. The cohort markedly differs from the general population in terms of sex distribution.6 The PAR% for employment in at least 1 at-risk occupation (global score) were 16% and 12% in men and women, respectively. For occupations related to specific industrial exposures, PAR% were in general higher in males (8% for asbestos, 9% for metals, 6% for PAH, 8% for silica) than in females (1% for asbestos, 1% for metals, <1% for PAH, 0% for silica). However the PAR% for ETS-related occupations was higher in women (6%) then in men (1%).
Asbestos, heavy metals and metal compounds, PAHs, silica, and ETS have been widely investigated in previous studies as risk factors for lung cancer. The contribution of the present study resides mainly in 2 aspects. First, we describe the results of a large population-based prospective study, with 217,055 participants followed for 6 years and 809 incident cases. As the information about previous occupational history was collected well before the occurrence of the tumors, recall bias can be ruled out. Questions about occupational history, although not including the time spent in each job, covered a large range of occupations, probably including most of known potentially harmful exposures for lung cancer. Many subjects held more than 1 (and sometimes several) occupations, with a complex pattern of hazardous exposures. Among the occupations associated with a substantial risk increase, we highlighted employment as a butcher (with a HR of 2.2 among women). Several studies have previously described the association between lung cancer and employment as a butcher or in slaughterhouses, but the results were not consistent, with some authors reporting a risk increase and others not.14–20 Employment as a butcher might involve exposure to PAHs, nitrosamines or other chemical agents, or exposure to animal viruses; we are still at the stage of speculation.
The second contribution of this study is in that we tried to simplify and summarize the exposures by classifying the occupations into 5 groups according to the main carcinogens potentially present. We created 5 different scores (for asbestos, metals, PAH, ETS, and silica) by summing all the occupations in which each subject was engaged. This procedure enabled us to gain statistical power and to tackle the problem of the inflation of type I error by limiting the number of hypotheses tested.
Among the subjects employed in any of the identified hazardous occupations, lung cancer incidence was 40% higher than in those employed elsewhere. Incidence tended to increase with the number of hazardous jobs reported (Fig. 1). This is probably related to the global time of exposure, but may also be due to the synergistic effect of different substances. Moreover, when the occupations were classified and grouped according to the suspected exposure to specific carcinogenic agents (ie, asbestos, metals, PAH, ETS) the HRs were usually larger, with a risk increase ranging from 40 to 100%. In some cases, the risks for women were even larger than for men.
All these results were adjusted for potential confounders, including country, age, sex, social class, diet, physical activity, and smoking habits. Only 54 incident cases out of 845 occurred among never-smokers. However, when we stratified current and former smokers for intensity and duration, the HRs for all the scores were reasonably stable across categories, suggesting that the confounding effect of smoking is likely to be modest. No significant interactions between smoke and employment were found, indicating a small or absent effect modification.
When we computed the PAR, we found that 16% of all lung cancer cases in men and 12% in women could be ascribed to professional exposures in one or more of the 52 selected occupations. Asbestos exposure alone could account for almost 8% of all cases in males. Among women, the proportion attributable to industrial exposures is less important, mainly due to a lower proportion of women involved in the relevant occupations, but 6% of all cases could be ascribed to occupations with ETS exposure. We could not account for total exposure to passive smoking in working places; our results are limited to employment in selected high-risk occupations (such as working in restaurants and bars).
When we compare these figures with the estimates published in the 1980s and 1990s in Europe and the United States, it seems that the proportion of lung cancer attributable to industrial exposures has not declined much in the last 20 years. Simonato et al21 in 1988 reported PARs for lung cancer ranging from 2.4 to 40% due to occupational exposures; Vineis et al22 in the same year estimated an attributable risk for industrial exposures in the Unites States ranging from 3 to 17%. Albin et al23 in 1999 reported a conservative estimate of 10 to 20% for asbestos occupational exposure in Europe, and Merler et al24 in the same year reported estimates ranging from 9 to 48% for high risk occupations in Italy.
The observed differences among men and women are worth mentioning. Global and specific attributable risks were generally lower in women (with the important exception of ETS-related occupations) due to the fact that women, at least in our sample, were less involved in industrial occupations at risk. Even so, the HRs associated with single employments are frequently larger in women (for example for farming, agriculture, and transportation). If we exclude potential biases (for example, the information about past employment could be more precise in women than in men), one explanation could be that smoking is usually more important as a cause of lung cancer among men, both in terms of prevalence and of intensity (data not shown), and this could in part conceal the carcinogenic effect of occupational exposures.
Our study has also some limitations, as follows. The occupational history was assessed by simple questions concerning only the engagement of the subjects in a list of a priori hazardous occupations. This excluded the possibility of finding new occupational risk factors. Also, the potential exposure misclassification due to limited information is expected to force the HR estimates toward the null value, and therefore our conclusions are likely to be on the conservative side. Second, no attempt was made to investigate the time spent in each job, or the effects of age at employment or of time elapsed since quitting. Third, the limited number of cases among nonsmokers did not enable us to focus the analyses on this subsample, although we could control for potential confounding from smoking by careful adjustment. Finally, our classification of jobs according to common exposures may be inaccurate.
In conclusion, our large prospective European study suggests that past exposure to occupational lung carcinogens, including passive smoking, is still associated with moderate to large increases in risk, and a significant proportion of all incident lung cancer cases is still attributable to occupational exposures. Specific actions should be carried out to more fully understand the existing risks and to prevent hazardous occupational exposures in the future.
We are grateful to Roel Vermeulen (University of Utrecht, the Netherlands) and to Dario Mirabelli (CPO Piemonte, Turin, Italy) for thoughtful comments. Mortality data for the Netherlands were obtained from “Statistics Netherlands”.
1. International Agency for Research on Cancer. Monographs on the Evaluation of Carcinogenic Risks to Humans
. Lyon: IARC; 1971–2005.
2. Boffetta P. Epidemiology of environmental and occupational cancer. Oncogene
3. Doll R, Peto R. The causes of cancer: quantitative estimates of avoidable risks of cancer in the United States today. J Natl Cancer Inst
4. Vineis P, Simonato L. Proportion of lung and bladder cancers in males resulting from occupation: a systematic approach. Arch Environ Health
5. Steenland K, Burnett C, Lalich N, et al. Dying for work: the magnitude of US mortality from selected causes of death associated with occupation. Am J Ind Med
6. Riboli E, Hunt KJ, Slimani N, et al. European Prospective Investigation into Cancer and Nutrition (EPIC): study populations and data collection. Public Health Nutr
7. Vineis P, Airoldi L, Veglia F, et al. Environmental tobacco smoke and risk of respiratory cancer and chronic obstructive pulmonary disease in former smokers and never smokers in the EPIC prospective study. BMJ
8. Leffondré K, Abrahamowicz M, Siemiatycki J, et al. Modelling smoking history: a comparison of different approaches. Am J Epidemiol
9. Wacholder S, Chanock S, Garcia-Closas M, et al. Assessing the probability that a positive report is false: an approach for molecular epidemiology studies. J Natl Cancer Inst
10. Coughlin SS, Chiazze L Jr. Job-exposure matrices in epidemiologic research and medical surveillance. Occup Med
11. Kauppinen T, Toikkanen J, Pukkala E. From cross-tabulations to multipurpose exposure information systems: a new job-exposure matrix. Am J Ind Med
12. Mirabelli D, Kauppinen T. Occupational exposures to carcinogens in Italy: an update of CAREX database. Int J Occup Environ Health
13. Breslow NE, Day NE. Statistical Methods in Cancer Research
. Vol. 1. Lyon, France: IARC Scientific Publications; 1980;74–78.
14. Boffetta P, Gridley G, Gustavsson P, et al. Employment as butcher and cancer risk in a record-linkage study from Sweden. Cancer Causes Control
15. al-Ghamdi AA, Sanders CM, Keefe M, et al. Human papillomavirus DNA and TP53 mutations in lung cancers from butchers. Br J Cancer
16. Coggon D, Wield G. Mortality of butchers and cooks identified from the 1961 census of England and Wales. Occup Environ Med
17. Guberan E, Usel M, Raymond L, et al. Mortality and incidence of cancer among a cohort of self employed butchers from Geneva and their wives. Br J Ind Med
18. Kristensen TS, Lynge E. Lung cancer among butchers and slaughterhouse workers. Scand J Work Environ Health
19. Gustavsson P, Fellenius E, Hogstedt C. Possible causes of increased lung cancer incidence among butchers and slaughterhouse workers. Scand J Work Environ Health
20. Durusoy R, Boffetta P, Mannetje A, et al. Lung cancer risk and occupational exposure to meat and live animals. Int J Cancer
21. Simonato L, Vineis P, Fletcher AC. Estimates of the proportion of lung cancer attributable to occupational exposure. Carcinogenesis
22. Vineis P, Thomas T, Hayes RB, et al. Proportion of lung cancers in males, due to occupation, in different areas of the USA. Int J Cancer
23. Albin M, Magnani C, Krstev S, et al. Asbestos and cancer: an overview of current trends in Europe. Environ Health Perspect
. 1999;107(suppl 2):289–298.
24. Merler E, Vineis P, Alhaique D, et al. Occupational cancer in Italy. Environ Health Perspect
. 1999;107(suppl 2):259–271.