Offermans, Nadine S.M. MSc; Vermeulen, Roel; Burdorf, Alex; Goldbohm, R. Alexandra; Kauppinen, Timo; Kromhout, Hans; van den Brandt, Piet A.
* Become familiar with previous studies on occupational asbestos exposure and risk of pleural mesothelioma, lung, and laryngeal cancer, with special attention to data gaps.
* Summarize the new findings on asbestos-related risks of pleural mesothelioma, lung, and laryngeal cancer specifically addressing risk associated with the lower end of the exposure distribution, risk of cancer subtypes, and the interaction between asbestos and smoking.
* Discuss the study implications, including job-exposure matrix-based exposure assessment and implications for calculating population-attributable fractions.
In 2006, globally, an estimated 125 million people were still occupationally exposed to asbestos1 with its use even increasing in parts of Asia, South America, and the former Soviet Union.2 In the Netherlands, despite being banned in 1993, asbestos is still a public health concern with respect to asbestos removal and site cleaning in the general environment.3 Asbestos research has been ongoing for decades and evidence has been accumulated that, regardless of fiber type, asbestos causes mesothelioma and lung, laryngeal, and ovarian cancer.4,5 Nevertheless, there remain questions around asbestos carcinogenicity that pertain to risk at the lower end of the exposure distribution,5–7 the possibility of uncontrolled confounding due to smoking and drinking for especially laryngeal cancer,8 the association with subtypes of lung9 and laryngeal cancer,8 and the interaction between asbestos and smoking in relation to lung10 and laryngeal cancer.8 Population-based studies are well suited to address these questions given their overall wide range in exposure levels, including those at the lower end of the exposure distribution (ie, exposure levels in jobs outside asbestos mining, insulation, cement and textile manufacturing, and other more highly exposed jobs), the possibility to control for potential confounders, and large size. The prospective Netherlands Cohort Study (NLCS) started in 1986 among 120,852 men and women of the general population.11,12 Besides questions on occupational history, the NLCS contains extensive information on dietary habits and lifestyle factors. Given the large study size and long follow-up, many cases of asbestos-related cancer have emerged in the NLCS. Therefore, the primary objectives of this study were to assess the following:
1. The overall association between occupational asbestos exposure and the risk of pleural mesothelioma, lung cancer, and laryngeal cancer, with special attention to risk associated with the lower end of the exposure distribution and to potential confounding
2. The association between occupational asbestos exposure and subtypes of lung and laryngeal cancer
3. The possible additive or multiplicative interaction between smoking and asbestos in relation to pleural mesothelioma, lung cancer, and laryngeal cancer
Because the proportion of long-term employed women was rather low, this study was conducted only among men.
We previously evaluated several methodologies for retrospective occupational exposure assessment within the NLCS. The job-exposure matrices (JEMs) DOMJEM and FINJEM showed rather similar agreement with case-by-case expert assessment and showed moderate agreement among each other.13 To provide insight into the methodological uncertainty associated with the choice of JEM, we present the main risk analyses using both DOMJEM and FINJEM.
MATERIAL AND METHODS
Study Population and Cancer Follow-Up
The study design and data collection strategies for the NLCS have been described in detail previously.11 In brief, the NLCS started in September 1986 when 58,279 men and 62,573 women aged 55 to 69 years, originating from 204 municipalities in the Netherlands with computerized population registries, were enrolled in the cohort. At baseline, participants completed a self-administered questionnaire on dietary habits and lifestyle, occupational history, and other potential risk factors for cancer.11 For reasons of efficiency in questionnaire processing and follow-up, the case–cohort approach was used.14 Incident cases were enumerated from the entire cohort, whereas the accumulated person-years at risk in the entire cohort were estimated from a random subcohort of 5000 subjects (2411 men and 2589 women), selected immediately after baseline. This subcohort is being followed up for vital status information, while the entire cohort is being monitored for incident cancer by annual record linkage to the Netherlands Cancer Registry and the Dutch Pathology Registry (PALGA).15,16 For these analyses, a total of 17.3 years of follow-up (baseline to December 2003) was available. Completeness of incident cancer coverage was estimated to be almost 100%.17 The NLCS was approved by the institutional review boards of the Netherlands Organisation for Applied Scientific Research TNO (Zeist) and Maastricht University (Maastricht).
All prevalent cases at baseline other than skin cancer were excluded, leaving 2336 male subcohort members, 160 pleural mesothelioma cases (International Classification of Diseases for Oncology, Third Edition [ICD-O-3] code C384), 2932 lung cancer cases (ICD-O-3 code C34), and 216 laryngeal cancer cases (ICD-O-3 codes C32.0 and C32.1, which refer to cancer of the glottis and supraglottis, respectively). The reason for considering only pleural mesothelioma and cancer of the glottis and supraglottis is the very low number of cases for peritoneal mesothelioma (n = 10) and cancer of the subglottis (n = 4).
Subjects without any, or only uncodable, information on occupational history or who never worked professionally were omitted from the analyses. As a result, 2107 male subcohort members, 145 pleural mesothelioma cases, 2592 lung cancer cases, and 184 laryngeal cancer cases were available for analyses after 17.3 years of follow-up.
Occupational Exposure Assessment
Information on lifetime occupational history until 1986 was obtained from the questionnaire completed at study enrolment. Questions concerned the job title, name and type of the company, products made in the department, and period of employment. On the basis of these questions, occupations were coded according to the Standard Occupational Classification of 1984 of the Dutch Central Bureau of Statistics, supplemented by a three-digit code assigned within the NLCS based on the job title. Subjects could enter a maximum of five occupations, which was generally sufficient to cover the lifetime occupational history for the large majority of the cohort, because cohort subjects held on average 1.9 job codes during their working life up to 1986. For all subjects, the job code was assessed for each of the maximally five occupations held between starting work and 1986.
We applied two JEMs, the DOMJEM from the Netherlands and the Finnish FINJEM, as described previously.13 Briefly, DOMJEM is a generic JEM developed by occupational exposure experts in the Netherlands for application in general population studies. It contains a combined measure of the probability × intensity of exposure, which is semiquantitative (no, low, or high exposure) with a weighting of 0, 1, and 4, respectively.18 FINJEM was constructed for exposure assessment in large register-based studies, is based on both expert assessment and exposure measurements, and contains a time axis.19 Although FINJEM was constructed for Finland, exposure estimates were not adapted to Dutch occupational circumstances before applying it in the NLCS.
Asbestos Exposure Variables
Several exposure variables were defined: ever versus never exposed to asbestos (yes/no), duration of exposure (years), cumulative exposure (CE; fiber-years/mL [f-y/mL] [FINJEM] or unit-years [DOMJEM], see explanation hereafter), and duration of high exposure (years). The CE is a combined measure of the probability (P), intensity (I), and duration (years) of exposure.
Ever versus never exposed is based on the CE, in that subjects were classified as being ever exposed to asbestos when CE > 0. For occupations with P × I > 0, duration of employment was summed to obtain the duration of exposure.
For DOMJEM, the CE measure was estimated by summing the product of P × I and duration over the reported occupations. The DOMJEM scores of no, low, and high exposure for P × I were arbitrarily assigned values of 0, 1, and 4, respectively, to mirror the log-normal (multiplicative) nature of occupational exposure levels, hence the expression in unit-years. The weighting was based on reported levels for semiquantitatively scored exposure, thereby ensuring a balanced weighting between intensity and duration in the calculation of CE.20
To arrive at the CE for FINJEM, first the P × I per job code was estimated using the time-specific exposure information in the time axis of this JEM, before summing P × I over the reported occupations. For those workers who started working before 1945, exposure was set to zero, because there was hardly any asbestos industry in the Netherlands in the period before 1945.
For the duration of high exposure, first, the P × I per occupation was categorized into no, low, or high exposure on the basis of the distribution in the subcohort. Second, duration of employment was summed for those occupations with a high P × I to obtain the duration of high exposure.
Participants were classified into never-exposed subjects and tertiles of those exposed to asbestos on the basis of the distribution among the subcohort for the duration of exposure and CE (reference group is the never exposed) and for the duration of high exposure (reference group is the never highly exposed). Continuous variables were also used. For the duration of (high) exposure, an increment of 10 years was used; while for the CE, an increment of 1 f-y/mL (FINJEM) or unit-year (DOMJEM) was used.
Age-adjusted (for lung cancer, also adjusted for family history of lung cancer [yes/no]) and multivariable-adjusted hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were estimated by using Cox proportional hazards (PH) models. Because age is the natural time scale in studies of disease occurrence, attained age was used as date of entry in the analyses.21 In models with (attained) age as date of entry, baseline age should also be considered a covariate when there is a cohort effect or when the proportional hazards assumption is violated,22 which was the case in our analyses. Because most PH assumptions was no longer violated after adjustment for baseline age, we included baseline age as a covariate in our analyses. The total person-years at risk were estimated from the subcohort,23 and standard errors were estimated by using a robust covariance matrix estimator to account for increased variance due to sampling the subcohort from the entire cohort.24
The covariates included in the multivariate models were either a priori selected risk factors based on the literature or variables that changed the age-adjusted regression coefficients by at least 10% (using a backward stepwise procedure). For mesothelioma, only current smoking (yes/no), the number of cigarettes smoked per day, and years of smoking cigarettes were considered potential confounders. Although we can assume that mesothelioma is solely caused by asbestos, we added these covariates for confirmation purposes only.
For lung cancer, the full covariate model consisted of current smoking (yes/no), the number of cigarettes smoked per day, years of smoking cigarettes, and occupational exposure to crystalline silica and polycyclic aromatic hydrocarbons. For laryngeal cancer, alcohol consumption was entered in the model in addition to current smoking (yes/no), the number of cigarettes smoked per day, and years of smoking cigarettes. Risk factors that were added to the model but did not satisfy the 10% rule and were not mentioned earlier were fruits and vegetables for all cancers and nickel, chromium, and welding fumes for lung cancer.
All covariates were entered into the models as continuous variables, except for current smoking (yes/no). The number of cigarettes smoked per day and years of smoking cigarettes were added to the model as both centered and noncentered variables because of possible problems with collinearity. As results of both analyses were comparable (not shown), we only presented the noncentered variables. To enable comparison, the age-adjusted (for mesothelioma and laryngeal cancer) and family history of lung cancer–adjusted (for lung cancer) analyses were restricted to subjects included in the multivariable-adjusted analyses (ie, with no missing values on confounding variables), which left 1962 subcohort members and 132 cases for pleural mesothelioma, 1962 subcohort members and 2324 cases for lung cancer, and 1931 subcohort members and 166 cases for laryngeal cancer for analyses.
For each analysis, the proportional hazards assumption was tested by using the scaled Schoenfeld residuals.25 Trends for all subjects and only including the exposed were evaluated with the Wald test by assigning subjects the median value for each level of the categorical variable among the subcohort members, and this variable was entered as a continuous term in the Cox regression model. In addition, we evaluated the association with subtypes of lung cancer (n = 379 for small cell lung cancer, 350 for large cell, 931 for squamous cell carcinoma, and 493 for adenocarcinoma) and laryngeal cancer (n = 122 for glottis and 44 for supraglottis).
Furthermore, we studied the interaction between asbestos exposure (yes/no) and smoking (no/former/current) for all three cancers both on a multiplicative and an additive scale. First, we examined whether the joint effect of asbestos and smoking was closer to additivity or multiplicativity for each cancer endpoint. Second, we tested for statistically significant departure from multiplicativity by including an interaction term in the Cox regression model. Third, we tested for statistically significant departure from additivity using the CI of the relative excess risk of cancer due to interaction, according to methods described by Knol and VanderWeele,26 which were adapted for use in Stata (Stata Corporation, College Station, TX). All analyses were performed using the Stata statistical software package (intercooled Stata, version 10). All tests were two-tailed, and differences were regarded as statistically significant at P < 0.05.
Finally, population-attributable fractions (PAFs) were calculated for all three cancers on the basis of the HRs for all asbestos exposure variables in the multivariable-adjusted models, using the formula: Σpi(RRi − 1)/[Σpi(RRi − 1) + 1], where p is the fraction of exposed subjects in the subcohort, i indexes the exposure level, and RR values are based on HRs.27
The distribution of asbestos exposure and potential confounders among male subcohort members and cancer cases in the NLCS is given in Table 1. Lung cancer cases more often reported a family history of lung cancer. In addition, they more often smoked cigarettes, and smoked more cigarettes per day and for more years than subcohort members, as did the laryngeal cancer cases. Cases for all three cancers consumed more alcohol per day and were, on average, more often and longer (highly) exposed to asbestos. Although FINJEM generally revealed a somewhat lower percentage of ever-exposed subjects, DOMJEM showed the lowest percentage of highly exposed subjects.
When studying the association between asbestos exposure and covariates in the subcohort (Table 2), subjects exposed to asbestos more often smoked cigarettes and for more years, with the JEMs showing different patterns over the tertiles. Alcohol consumption was higher for those never exposed to asbestos than for those exposed, as classified according to both JEMs. There was an association between exposure to silica and polycyclic aromatic hydrocarbons and asbestos exposure, with DOMJEM showing an increasing pattern over the tertiles of exposure, while the pattern for FINJEM was less clear.
Overall, all asbestos exposure variables were positively associated with risk of mesothelioma, lung cancer, and laryngeal cancer, using both DOMJEM and FINJEM, and for the age-adjusted (mesothelioma and laryngeal cancer) and for family history of lung cancer–adjusted (yes/no) (lung cancer) models as well as the multivariable-adjusted models (Tables 3 to 5). Adjusting for potential confounders had no influence on the association with mesothelioma, and little to no effect on the association with lung and laryngeal cancer. Therefore, only multivariable-adjusted results are mentioned in the text hereafter.
Overall, DOMJEM revealed a higher HR (95% CI) for the duration of high exposure (tertile 3 vs never: HR = 13.66 [95% CI, 5.86 to 31.84] for mesothelioma; HR = 2.99 [95% CI, 1.39 to 6.41] for lung cancer; and HR = 6.36 [95% CI, 2.18 to 18.53] for laryngeal cancer) than FINJEM (HR = 3.28 [95% CI, 1.82 to 5.92] for mesothelioma; HR = 1.74 [95% CI, 1.20 to 2.54] for lung cancer; and HR = 1.49 [95% CI, 0.75 to 2.97] for laryngeal cancer).
FINJEM showed a higher HR (95% CI) for ever versus never exposed (HR = 3.02 [95% CI, 2.11 to 4.34] for mesothelioma; HR = 1.50 [95% CI, 1.27 to 1.78] for lung cancer; and HR = 1.42 [95% CI, 0.99 to 2.03] for laryngeal cancer) than DOMJEM (HR = 2.62 [95% CI, 1.82 to 3.76] for mesothelioma; HR = 1.19 [95% CI, 1.02 to 1.40] for lung cancer; and HR = 1.20 [95% CI, 0.84 to 1.72] for laryngeal cancer).
For mesothelioma (Table 3), associations generally reached statistical significance and showed a clear dose–response relation when including the never-exposed subjects (Ptrend < 0.001). When only including the exposed subjects, only DOMJEM showed a significant trend (Ptrend < 0.001).
For lung cancer (Table 4), not all associations were statistically significant, though tests for trend were significant when including the never-exposed subjects (Ptrend < 0.05). When only the exposed subjects were considered, trends were no longer significant. Results by histology of lung cancer were fairly comparable to overall lung cancer apart from adenocarcinoma, for which associations with most exposure variables were weaker or absent. The only exception was the duration of high exposure for DOMJEM, which showed a statistically significant association for the continuous variable (HR = 1.43 [95% CI, 1.06 to 1.93]) and positive trend (Ptrend = 0.047) for risk of adenocarcinoma.
TABLE 4-a. Hazard Ra...Image Tools
For laryngeal cancer (Table 5), risk by location showed usually stronger associations for supraglottis than glottis cancer, except for the duration of high exposure for DOMJEM, which was statistically significant for both the highest tertile (HR = 7.09 [95% CI, 2.31 to 21.74]) and the continuous variable, as was the dose–response relation (Ptrend = 0.002).
The joint effect of asbestos and smoking for each cancer endpoint was assessed by comparing the HRs of current smokers not exposed to asbestos and never-smokers exposed to asbestos with the HR of current smokers exposed to asbestos (Table 6). Given space limitations and because HRs for ever versus never exposed were overall higher for FINJEM than for DOMJEM, we will only present results for FINJEM.
For mesothelioma, the observed HR of 4.18 for the combined exposure category is lower than the product of HRs for asbestos and smoking (5.39 × 1.35 = 7.28), indicating that the joint effect is less than multiplicative. The observed HR is also lower than expected when using an additive model (5.39 + 1.35 − 1 = 5.74). For lung cancer, the observed HR of 10.21 is lower than the product of asbestos and smoking (7.48 × 1.79 = 13.39), indicating that the joint effect is less than multiplicative. The observed HR is higher than what would have been expected by using an additive model (7.48 + 1.79 − 1 = 8.27). For laryngeal cancer, the observed HR of 20.73 is lower than the product (16.95 × 2.27 = 38.48), indicating that the joint effect is considerably less than multiplicative. The observed HR is somewhat higher than what would have been expected by using an additive model (16.95 + 2.27 − 1 = 18.22).
When testing for departure from multiplicativity (by including an interaction term) or additivity (using the CI of relative excess risk of cancer due to interaction), there was no statistically significant multiplicative or additive interaction between asbestos and smoking for any of the cancers. Relative excess risk of cancer due to interaction and 95% CIs are presented in Table 6.
Population-attributable fractions were highest for mesothelioma and ranged from 15.9% to 34.5%, depending on the asbestos exposure variable and JEM used. Population-attributable fractions for lung and laryngeal cancer were more or less comparable and were in the range of 2.5% to 12.1% for lung cancer and 2.9% to 10.0% for laryngeal cancer (Tables 3 to 5).
This study was able to confirm the well-established associations between asbestos exposure and mesothelioma, lung cancer, and laryngeal cancer, revealing elevated risks at the lower end of the exposure distribution in the population-based NLCS. Associations with lung cancer subtypes were generally comparable to overall lung cancer, except for adenocarcinoma, which showed only a weak positive association after prolonged higher asbestos exposure. For laryngeal cancer, associations were usually stronger for supraglottis than glottis cancer, which showed only a positive association after prolonged higher asbestos exposure. There was no statistically significant interaction on an additive or multiplicative scale between asbestos and smoking in relation to pleural mesothelioma, lung cancer, and laryngeal cancer. The estimated joint effect of asbestos and smoking was both lower than additive and multiplicative for mesothelioma. For lung cancer, the joint effect was between additivity and multiplicativity, while for laryngeal cancer the joint effect was closer to additivity than multiplicativity. Finally, results were robust against the use of different JEMs, though DOMJEM showed higher HRs for the duration of high exposure and FINJEM revealed somewhat higher HRs for ever versus never exposed.
The observed positive associations between asbestos and mesothelioma, lung cancer, and laryngeal cancer in the NLCS are qualitatively comparable to many other studies.5,8 Quantitatively, results are more difficult to compare, because the range in relative risks reported in the literature is wide, certainly for lung cancer and mesothelioma, which have been studied extensively. Furthermore, the NLCS is a population-based study with a wide range in exposure levels, including those at the lower end of the exposure distribution. Low asbestos exposure levels are currently present in most industrialized countries, and the magnitude of the cancer risks associated with them is of importance for setting acceptable exposure limits.
For mesothelioma, HRs were significantly elevated in this study, even for the lowest tertile of CE (median, 0.20 f-y/mL) based on FINJEM (HR = 2.69 [95% CI, 1.60 to 4.53]). When using DOMJEM, no increased HR was observed for the lowest tertile. As the time since first exposure to asbestos was more than 20 years for all but one worker with mesothelioma, the length of follow-up was long enough for mesothelioma to develop. Some previous studies found no evidence of a threshold level for asbestos-related mesothelioma,28,29 though others did find evidence of a threshold level.30,31 Because there is no uniform definition of what low level exposure entails, and because exposure assessment in our study is JEM-based possibly entailing nondifferential exposure misclassification, we cannot subscribe on the basis of our data to the (non)existence of a threshold level for asbestos-related mesothelioma.
For lung cancer, this study showed a significantly increased HR even for the lowest tertile of CE based on FINJEM (HR = 1.47 [95% CI, 1.15 to 1.87]). Again, when using DOMJEM, no increased HR was observed for the lowest tertile. The FINJEM estimate was comparable to or higher than the relative risks presented in the meta-analysis by van der Bij et al,32 depending on the model they used, for a higher exposure level of 4 f-y/mL. In this meta-analysis, they proposed that the asbestos-related increase in relative risk of lung cancer may be larger than expected from previous meta-analyses.32 Also, Gustavsson et al6 found, in their population-based study, a higher relative risk of lung cancer at the lower end of the exposure distribution than predicted by downward linear extrapolation from highly exposed occupational cohorts. Possibly, relative risks at the lower end of the exposure distribution are higher than expected.
Laryngeal cancer has not been studied as extensively in relation to asbestos as mesothelioma and lung cancer and has only recently been linked causally to asbestos.4 For FINJEM, our results were comparable to the summary relative risk of 1.43 for ever versus never exposed to asbestos in a meta-analysis of case–control studies, while for DOMJEM, the HR for ever versus never exposed was slightly lower.8 Adjusting for alcohol consumption and smoking did not alter these results, which is in line with several previous studies on asbestos and laryngeal cancer.8
Population-attributable fractions discussed hereafter are based on ever versus never exposed for ease of comparison with other studies. In the NLCS, PAFs for mesothelioma (31.9% for DOMJEM and 34.3% for FINJEM) are lower than reported in the literature.33–35 Because virtually all mesothelioma is due to (occupational) asbestos, PAFs of around 100% might have been expected. The NLCS is, however, a population-based study with a wide range in exposure levels and a modest exposure prevalence, which both determine the PAF. Moreover, exposure assessment using JEMs could entail nondifferential exposure misclassification leading to attenuated HRs and consequently PAFs. Indeed, the percentage never exposed according to both JEMs is around 50%. If we assume mesothelioma to be solely caused by asbestos, this percentage of 50% may point to nonoccupational asbestos exposure but probably also to incidental occupational asbestos exposures in jobs not captured by the JEMs.
Population-attributable fractions for lung and laryngeal cancer were only slightly lower than reported in the literature. For lung cancer, Albin et al33 reported PAFs in the range of 10% to 20%, though other studies found lower estimates.36,37 An earlier publication in the NLCS found a PAF of 11.6%,12 which is comparable to the PAF for FINJEM (11.5%) in the present study and twice as high as the PAF for DOMJEM (5.2%). PAFs in this earlier study were, however, based on case-by-case expert assessment, a shorter follow-up, and the probability of exposure (without including duration). For laryngeal cancer, the PAF of 8.3% calculated by Nurminen and Karjalainen35 was higher than the PAF of 5.5% for DOMJEM, but lower than the PAF of 9.8% for FINJEM in the NLCS.
This study indicated adenocarcinoma to be the lung cancer subtype showing a weak association only after prolonged higher asbestos exposure. The weaker association between asbestos and adenocarcinoma is of interest, because previous studies showed asbestos to be associated preferentially with adenocarcinoma,38,39 though this association has not been reported consistently throughout the literature.40,41 Because asbestos and smoking may have similar exposure routes and related pathophysiological pathways,5 and because smoking is only weakly associated with adenocarcinoma,40,41 this might explain why only after prolonged higher asbestos exposure a positive association with adenocarcinoma was observed. Another reason might be that adenocarcinoma is more frequently observed among nonsmokers,40 and because most subjects exposed to asbestos were smokers, the true association between asbestos and adenocarcinoma may have been stronger. When stratifying by smoking, the association with adenocarcinoma for never-smokers ever exposed to asbestos compared with never-smokers never exposed to asbestos was indeed higher (HR = 2.43 [95% CI, 0.83 to 7.11]), though numbers were low (FINJEM results, not shown).
Relative risk by location of laryngeal cancer in the NLCS is comparable to some, though not all studies.8 The NLCS showed overall stronger associations for supraglottis than glottis cancer. Because the location of the supraglottis seems more readily exposed to tobacco,42 and because exposure routes and pathophysiological pathways of tobacco and asbestos are believed to be related,5 the higher HR for cancer of the supraglottis might have been expected. Unfortunately, we could not check whether and to what degree this association between asbestos and cancer of the supraglottis was driven by smoking, because all asbestos-exposed men were smoking. In addition, the numbers were low, certainly for supraglottis cancer.
As expected, this study showed that the HR of mesothelioma was not influenced by smoking as opposed to HRs of lung and laryngeal cancer. Although smoking is paramount in the asbestos–lung cancer association, smoking is no risk factor for mesothelioma.43 A synergistic effect between asbestos and smoking in relation to lung cancer is supported by a number of systematic reviews,44–46 though the degree of synergism remains uncertain because results range from additivity to supramultiplicativity.6,10 Our study found no statistically significant interaction on an additive or multiplicative scale for lung cancer, and joint effects were between additivity and multiplicativity.
For laryngeal cancer, there was also no statistically significant interaction on an additive or multiplicative scale, though joint effects were closer to additivity than multiplicativity. Results have to be interpreted carefully, as the numbers were low most notably among reference category of the never-smokers not exposed to asbestos. Previous studies also observed a joint effect closer to additivity than multiplicativity, though one study found a supramultiplicative joint effect of asbestos and smoking.8
This study had the advantage of using two JEMs for studying asbestos–cancer associations to get insight into the methodological uncertainty associated with the choice of JEM. Overall, there was a high degree of similarity in results between both JEMs, which reinforced our confidence in the results presented. Nevertheless, both JEMs may have their particular strengths. Possibly, DOMJEM is better in selecting the prolonged highly exposed subjects as reflected by higher HRs for the duration of high exposure and a lower exposure prevalence of the ever highly exposed. FINJEM may be better in discriminating between ever and never exposed, as reflected by somewhat higher HRs for ever versus never exposed and a lower exposure prevalence of the ever exposed. Furthermore, PAFs were affected by this particular difference between both JEMs and were overall higher (up to threefold) for FINJEM than for DOMJEM. Therefore, some caution seems appropriate when judging PAFs that are JEM-based.
Because exposure assessment in population-based studies like the NLCS has to be based on rather brief occupational histories and JEMs, nondifferential exposure misclassification could have biased the exposure–response relationships. Both JEMs were, however, able to confirm the well-known association between asbestos exposure and mesothelioma.
Strengths of this study included the prospective design; the long, nearly complete follow-up and large study size; and the possibility to correct for several lifestyle confounders such as smoking and alcohol. The prospective design reduced the potential for recall bias, most notably on potential confounders as smoking and alcohol, and the nearly complete follow-up of cases and subcohort members made selection bias unlikely. The length of follow-up ensured a long-enough latency to develop mesothelioma and enough power to study subtypes and interaction between asbestos and smoking. Finally, the NLCS is a population-based study with a wide range in exposure levels, including those at the lower end of the exposure distribution, which are nowadays present in most industrialized countries.
This study was able to confirm the well-established associations between asbestos exposure and mesothelioma, lung cancer, and laryngeal cancer, revealing elevated risks at the lower end of the exposure distribution in the population-based NLCS, also after adjusting for several lifestyle confounders. Relative risks by lung cancer histology are comparable to overall lung cancer results, except for adenocarcinoma, which showed only a weak association after prolonged higher asbestos exposure. Supraglottis cancer seems to have a stronger association with asbestos exposure than glottis cancer. There was no evidence of statistically significant additive or multiplicative interaction between asbestos and smoking in relation to pleural mesothelioma, lung cancer, and laryngeal cancer. Nevertheless, joint effects of asbestos and smoking were closer to additivity than multiplicativity for laryngeal cancer. Finally, results were robust against the use of different JEMs, with DOMJEM and FINJEM resulting in essentially similar HRs.
The authors are indebted to the participants of this study, the Netherlands Cancer Registry, and the Dutch Pathology Registry (PALGA). They also thank S. van de Crommert, Dr L. Schouten, J. Nelissen, C. de Zwart, M. Moll, S. van den Heuvel, and A. Pisters for their assistance with data entry and/or data management; L. Preller regarding use of FINJEM; S. Peters regarding use of DOMJEM; A. Volovics and A. Kester for statistical advice; and H. van Montfort, T. van Moergastel, E. Dutman, and R. Schmeitz for programming assistance.
2. LaDou J. The asbestos cancer epidemic. Environ Health Perspect. 2004;112:285–290.
4. Straif K, Benbrahim-Tallaa L, Baan R, et al. A review of human carcinogens—part C: metals, arsenic, dusts, and fibres. Lancet Oncol. 2009;10:453–454.
5. International Agency for Research on Cancer. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans: Asbestos (Chrysotile, Amosite, Crocidolite, Tremolite, Actinolite and Anthophyllite). Lyon, France: International Agency for Research on Cancer; 2012.
6. Gustavsson P, Nyberg F, Pershagen G, Scheele P, Jakobsson R, Plato N. Low-dose exposure to asbestos and lung cancer: dose–response relations and interaction with smoking in a population-based case-referent study in Stockholm, Sweden. Am J Epidemiol. 2002;155:1016–1022.
7. Ogden TL. Canadian chrysotile report released—at last. Ann Occup Hyg. 2009;53:307–309.
8. NAS. Asbestos: Selected Cancers: The National Academies, Institute of Medicine, Board on Population Health and Public Health Practices, Committee on Asbestos: Selected Health Effects. Washington, DC: The National Academies Press; 2006:173–192.
9. Gonzalez M, Vignaud JM, Clement-Duchene C, et al. Smoking, occupational risk factors, and bronchial tumor location: a possible impact for lung cancer computed tomography scan screening. J Thorac Oncol. 2012;7:128–136.
10. International Agency for Research on Cancer. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans: Tobacco Smoke: Lyon, France: International Agency for Research on Cancer; 2004:917–919.
11. van denBrandt PA, Goldbohm RA, van ‘t Veer P, Volovics A, Hermus RJ, Sturmans F. A large-scale prospective cohort study on diet and cancer in the Netherlands. J Clin Epidemiol. 1990;43:285–295.
12. van Loon AJ, Kant IJ, Swaen GM, Goldbohm RA, Kremer AM, van den Brandt PA. Occupational exposure to carcinogens and risk of lung cancer: results from the Netherlands Cohort Study. Occup Environ Med. 1997;54:817–824.
13. Offermans NS, Vermeulen R, Burdorf A, et al. Comparison of expert and job-exposure matrix-based retrospective exposure assessment of occupational carcinogens in the Netherlands Cohort Study. Occup Environ Med. 2012;69:745–751.
14. Prentice RL. A case–cohort design for epidemiologic cohort studies and disease prevention trials. Biometrika. 1986;73:1–11.
15. Casparie M, Tiebosch AT, Burger G, et al. Pathology databanking and biobanking in the Netherlands, a central role for PALGA, the nationwide histopathology and cytopathology data network and archive. Cell Oncol. 2007;29:19–24.
16. Van den Brandt PA, Schouten LJ, Goldbohm RA, Dorant E, Hunen PM. Development of a record linkage protocol for use in the Dutch Cancer Registry for Epidemiological Research. Int J Epidemiol. 1990;19:553–558.
17. Goldbohm RA, van den Brandt PA, Dorant E. Estimation of the coverage of Dutch municipalities by cancer registries and PALGA based on hospital discharge data. Tijdschr Soc Gezonheidsz. 1994;72:80–84.
18. Peters S, Vermeulen R, Cassidy A, et al. Comparison of exposure assessment methods for occupational carcinogens in a multi-centre lung cancer case–control study. Occup Environ Med. 2011;68:148–153.
19. Kauppinen T, Toikkanen J, Pukkala E. From cross-tabulations to multipurpose exposure information systems: a new job-exposure matrix. Am J Ind Med. 1998;33:409–417.
20. Stewart PA, Herrick RF, Blair A, et al. Highlights of the 1990 Leesburg, Virginia, International Workshop on Retrospective Exposure Assessment for Occupational Epidemiology Studies. Scand J Work Environ Health. 1991;17:281–285.
21. Korn EL, Graubard BI, Midthune D. Time-to-event analysis of longitudinal follow-up of a survey: choice of the time-scale. Am J Epidemiol. 1997;145:72–80.
22. Gail MH, Graubard B, Williamson DF, Flegal KM. Comments on “Choice of time scale and its effect on significance of predictors in longitudinal studies” by Michael J. Pencina, Martin G. Larson and Ralph B. D'Agostino, Statistics in Medicine 2007;26:1343–1359. Stat Med. 2009;28:1315–1317.
23. Volovics A, van den Brandt PA. Methods for the analysis of case–cohort studies. Biomed J. 1997;39:195–214.
24. Barlow WE, Ichikawa L, Rosner D, Izumi S. Analysis of case–cohort designs. J Clin Epidemiol. 1999;52:1165–1172.
25. Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika. 1982;69:239–241.
26. Knol MJ, VanderWeele TJ. Recommendations for presenting analyses of effect modification and interaction. Int J Epidemiol. 2012;41:514–520.
27. Steenland K, Armstrong B. An overview of methods for calculating the burden of disease due to specific risk factors. Epidemiology. 2006;17:512–519.
28. Iwatsubo Y, Pairon JC, Boutin C, et al. Pleural mesothelioma: dose–response relation at low levels of asbestos exposure in a French population-based case–control study. Am J Epidemiol. 1998;148:133–142.
29. Hillerdal G. Mesothelioma: cases associated with non-occupational and low dose exposures. Occup Environ Med. 1999;56:505–513.
30. Ilgren EB, Browne K. Asbestos-related mesothelioma: evidence for a threshold in animals and humans. Regul Toxicol Pharmacol. 1991;13:116–132.
31. Gibbs GW, Berry G. Mesothelioma and asbestos. Regul Toxicol Pharmacol. 2008;52:S223–S231.
32. van der Bij S, Koffijberg H, Lenters V, et al. Lung cancer risk at low cumulative asbestos exposure: meta-regression of the exposure–response relationship. Cancer Causes Control. 2013;24:1–12.
33. Albin M, Magnani C, Krstev S, Rapiti E, Shefer I. Asbestos and cancer: an overview of current trends in Europe. Environ Health Perspect. 1999;107(suppl 2):289–298.
34. Driscoll T, Nelson DI, Steenland K, et al. The global burden of disease due to occupational carcinogens. Am J Ind Med. 2005;48:419–431.
35. Nurminen M, Karjalainen A. Epidemiologic estimate of the proportion of fatalities related to occupational factors in Finland. Scand J Work Environ Health. 2001;27:161–213.
36. Nicholson WJ, Perkel G, Selikoff IJ. Occupational exposure to asbestos: population at risk and projected mortality—1980–2030. Am J Ind Med. 1982;3:259–311.
37. Morabia A, Markowitz S, Garibaldi K, Wynder EL. Lung cancer and occupation: results of a multicentre case–control study. Br J Ind Med. 1992;49:721–727.
38. Johansson L, Albin M, Jakobsson K, Mikoczy Z. Histological type of lung carcinoma in asbestos cement workers and matched controls. Br J Ind Med. 1992;49:626–630.
39. Raffn E, Lynge E, Korsgaard B. Incidence of lung cancer by histological type among asbestos cement workers in Denmark. Br J Ind Med. 1993;50:85–89.
40. Lee BW, Wain JC, Kelsey KT, Wiencke JK, Christiani DC. Association of cigarette smoking and asbestos exposure with location and histology of lung cancer. Am J Respir Crit Care Med. 1998;157:748–755.
41. Paris C, Benichou J, Saunier F, et al. Smoking status, occupational asbestos exposure and bronchial location of lung cancer. Lung Cancer. 2003;40:17–24.
42. Schottenfeld D, Fraumeni JF. Cancer Epidemiology and Prevention. New York, NY: Oxford University Press; 2006:627–637.
43. Kanarek MS. Mesothelioma from chrysotile asbestos: update. Ann Epidemiol. 2011;21:688–697.
44. Erren TC, Jacobsen M, Piekarski C. Synergy between asbestos and smoking on lung cancer risks. Epidemiology. 1999;10:405–411.
45. Lee PN. Relation between exposure to asbestos and smoking jointly and the risk of lung cancer. Occup Environ Med. 2001;58:145–153.
46. Liddell FD. The interaction of asbestos and smoking in lung cancer. Ann Occup Hyg. 2001;45:341–356.
47. Swuste P, Dahhan M, Burdorf A. Linking expert judgement and trends in occupational exposure into a job-exposure matrix for historical exposure to asbestos in the Netherlands. Ann Occup Hyg. 2008;52:397–403.
Copyright © 2014 by the American College of Occupational and Environmental Medicine