Article Text

Download PDFPDF

Occupational exposure to silica and lung cancer risk in the Netherlands
  1. Liesbeth Preller1,
  2. Linda M C van den Bosch1,
  3. Piet A van den Brandt2,
  4. T Kauppinen3,
  5. Alexandra Goldbohm1
  1. 1TNO Quality of Life, Zeist/Leiden, The Netherlands
  2. 2Department of Epidemiology, GROW School for Oncology and Developmental Biology, Maastricht University, Maastricht, The Netherlands
  3. 3Finnish Institute of Occupational Health, Helsinki, Finland
  1. Correspondence to Sandra Goldbohm, TNO Quality of Life/Prevention and Health, PO Box 2215, 2301 CE Leiden, The Netherlands; sandra.bausch{at}tno.nl

Abstract

Objectives The lung cancer carcinogenicity of crystalline silica dust remains the subject of discussion. Epidemiological evidence is based on occupational cohort studies and population-based case–control studies. The aim of this study was to assess associations between male lung cancer risk and silica exposure in a population-based cohort study.

Methods The study was conducted among men aged 55–69 years (n=58 279) from the Netherlands Cohort Study, which included self-reported, life-time job histories. Job titles were linked to the occupational groups of the external Finnish Job Exposure Matrix (FINJEM), including probability and level of silica exposure, each for specific time periods. 1667 incident lung cancer cases with known silica exposure status (210 exposed) were available after 11.3 years of follow-up. Risks were estimated based on a case–cohort design, and using Cox proportional hazards models.

Results Adjusted for smoking and other confounders, elevated risks were observed for exposure duration (RR 1.65, 95% CI 1.14 to 2.41 for 26–51 years vs no exposure) and cumulative exposure (RR 1.47, 95% CI 0.93 to 2.33 for ≥3 vs <3 mg/m3.year). Associations with average exposure levels were weaker. Associations were stronger for occupations with an exposure probability of ≥90%. Adjustment for asbestos exposure slightly increased the risk.

Conclusions Results from this prospective population-based cohort study corroborate the classification of crystalline silica as a lung carcinogen. Associations could not be explained by smoking or by asbestos exposure.

  • lung cancer
  • silica exposure
  • cohort study

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

What this paper adds

  • Industrial cohort studies and population-based case–control studies generally show an increased risk of lung cancer related to silica exposure.

  • Missing data on potential confounders or on exposure information as well as other study limitations, led to uncertainty in exposure–response associations.

  • This large prospective cohort study, including detailed information on individual working conditions and potential confounders, and using exposure data from an external job exposure matrix strongly shows an increased risk of lung cancer related to silica exposure.

Introduction

Crystalline silica is classified by IARC as carcinogenic to humans (class 1) because it is thought to cause lung cancer. Epidemiological evidence for this classification is, however, inconsistent. Meta-analyses show strong associations between silica exposure and lung cancer risk, with the highest risks found among patients with silicosis and no or slightly increased risks among those without silicosis.1–6 The potentially distorting role of pre-existing silicosis is as yet not fully understood.1 7 Whereas many individual studies as well as a pooled analysis of 10 independent studies show positive exposure–response relationships, not all do. In the positive studies, the magnitude of the associations differs across studies.8 Differences in the origin and physical properties of silica exposure, level of exposure and the prevalence of silicosis among the studied population may contribute to heterogeneity in exposure–response relationships between studies.8 9 Besides, silica is suspected of being a weak carcinogen, with risk estimates mostly below 2 for the highest exposed groups. Small numbers of cases and the large potential of non-differential misclassification of biologically relevant exposure hamper detection of true elevated exposure–response relationships. The often lacking information on smoking habits in industrial cohorts may lead to inaccurate risk estimates because of the prevailing role of smoking in lung cancer. Several studies indicated that not adjusting for smoking habits may overestimate the risk when considering the total group including smokers and non-smokers10–12 and smokers alone,13 but the risk seems to be underestimated for non-smokers in cohort studies.13

The Netherlands Cohort Study (NLCS), a large population-based prospective cohort study primarily focusing on associations between diet and cancer, has the important advantage of having individual data on many potential confounders, including smoking. It also contains information on occupational history. Earlier analyses within this cohort showed elevated lung cancer risks in industrial groups with a high probability of silica exposure: the construction industry and the building materials, glass, clay and stone production industry.14 We wanted to elaborate on these findings and see whether these associations could potentially be attributed to silica exposure. We therefore applied an external job exposure matrix (JEM), the Finnish Job Exposure Matrix (FINJEM),15 to the NLCS for assessment of exposure–response relationships. In FINJEM, which has been developed for application in population-based epidemiological studies in Finland, crystalline silica exposure levels have been estimated for distinct occupational groups, in different time frames. Use of these data in a Finnish population showed positive associations between silica exposure and lung cancer.11

Methods

Study population

The NLCS started in September 1986 when 120 852 men and women aged 55–69 years from 204 municipalities were enrolled in the cohort using computerised population registries.16 This represented about 36% of the people invited to participate. Because the proportion of long-time employed women and the number of female lung cancer cases was quite small, the analyses were restricted to male lung cancer. At enrolment, the 58 279 men completed a self-administered mailed questionnaire on habitual dietary intake and other potential risk factors for cancer, such as smoking, lifestyle, education and occupational history. After the baseline exposure assessment, a subcohort of 5000 subjects (2411 men) was randomly sampled from the full cohort. Following the case–cohort design, this subcohort was followed for migration and vital status to estimate person time at risk accumulated in the full cohort. The sub cohort described in this paper was followed up for 11.3 years, from September 1986 to December 1997. Two male subcohort members were lost to follow up. The cohort at large has been followed-up for incident cancer by record linkage to the Netherlands Cancer Registry and the Netherlands Pathology Registry (PALGA) for 11.3 years.17 It has been shown that coverage of cancer incidence in the cohort is more than 95%.16 Prevalent cancer cases other than non-melanoma skin cancer were excluded from the subcohort. This left 2251 subcohort members and 1920 lung cancer cases. A total of 217 subcohort members and 253 lung cancer cases without any information on occupational history, or who had never worked professionally, were omitted from the analyses. As a result, after 11.3 years of follow-up, 1667 incident primary male lung carcinoma cases (ICD-O-3 code: C34) and 2034 subcohort members were available for analysis.

The NLCS was approved by the institutional review boards of both the Netherlands Organisation for Applied Scientific Research TNO (Zeist) and Maastricht University (Maastricht).

Occupational exposure assessment

Information on life-time occupational history until 1986 was obtained from the questionnaire filled in at study enrolment. Occupation was coded according to the Standard Occupational Classification (1984) of the Dutch Central Bureau of Statistics (CBS), using the self-reported information on job title, name and type of company, products made in the department and period of employment. Hence, occupation was assessed for each year between starting work and 1986. For each person, information on a maximum of five occupations was registered.

Since no JEM for silica exposure is available for the Dutch population, FINJEM15 was used to assign exposure information to individual workers. For each year of the total occupational history of each individual in the dataset, exposure was estimated using the time specific exposure information for occupational groups in FINJEM as indicated in table 1. Exposure information for the period 1945–1959 was also used for the period prior to 1945 for those workers who started working before 1945. For 10% of the population some information on occupational history could not be coded. In these cases, exposure was set to zero for the period with unclear exposure.

Table 1

FINJEM exposure information for occupational groups potentially exposed to silica and present within the study population

In FINJEM exposure to silica is considered to occur in 70 occupational subgroups. These 70 subgroups were linked by one of us (LP) to the corresponding Dutch occupations, out of the total of 14 000 listed in the Dutch Standard Occupational Classification. In case of doubt, classification was discussed with one of the other authors (TK). Linkage was carried out without knowledge of distribution of occupations over disease status. In FINJEM, the 70 subgroups were aggregated into 27 occupational groups (21 of which are present in our population) with distinct exposure levels and/or probability. Probability is defined as the percentage of workers estimated to be actually exposed within an occupational group, and level of silica exposure concentration is expressed in mg/m3 occupational inhalatory exposure to respirable (aerodynamic diameter <5 μm) quartz or crystalline silica containing dusts. Amorphous silica dust is not included. Within an occupational group, exposure was considered to be above zero when at least 5% of the workers were exposed to an annual average level of 0.02 mg/m3 of silica at any time in the period 1945–1995, with the exception of some groups for whom exposure was set to 0.01 mg/m3. For groups with lower or no exposure, exposure was set to zero.

Statistical analyses

Cox proportional hazards models were developed to estimate rate ratios (RRs) and 95% CIs relating exposure to the incidence of lung cancer (STCOX procedure, Stata version SE 8.0). The 95% CIs were corrected for the additional variance introduced by using person time estimates from a randomly sampled subcohort instead of the complete cohort,18 by using the robust option.

Associations were assessed for the following exposure variables: ever/never exposed (yes/no), duration of employment with silica exposure (years), average exposure concentration while working in exposed jobs (mg/m3) and cumulative exposure (mg/m3.year) during total occupational history. These exposure variables were not weighted for exposure probability. Exposure variables were analysed using a linear scale and a log transformed scale (average and cumulative exposure), and categorised based on the distribution among the subcohort (dichotomised or tertiles) and/or the current Dutch occupational exposure limit for exposure to 0.075 mg/m3 crystalline silica. As an alternative for cumulative exposure, duration and average concentration were also used simultaneously as independent variables in the models.

Analyses were performed on the entire population, which is defined here as all cases and all subcohort members available for analysis, as well as on a population with a high degree of certainty of the exposure status of each individual person. For the latter analyses, the population was restricted in two ways (hereafter called the ‘restricted’ population):

  1. Restriction of ‘never’ exposed: all those with any missing or uncodable data for any part of their occupational history were excluded from the analyses.

  2. Restriction of exposed:

    1. Men having held one or more jobs in occupational groups with ≥90% probability of silica exposure were included. For this group, exposure variables were calculated for the periods they worked in occupations with ≥90% probability of exposure; exposure during other periods, if any, was set to zero.

    2. Men having held one or more jobs in occupational groups with less than 90% probability of silica exposure and no jobs in occupational groups with ≥90% probability of silica exposure were excluded from the analyses. The latter was done as false positive exposure estimates have a large impact on results in general population studies.19

The restricted population contained 1426 lung cancer cases and 1723 subcohort members.

To test potential confounding, three models with different sets of potential confounders were used: 1 (basic model): age and family history of lung cancer; 2 (intermediate model): set 1 plus smoking behaviour (three variables simultaneously: current smoking status (yes/no), average number of cigarettes smoked per day, total years of smoking cigarettes); and 3 (full model): set 2 plus alcohol consumption (0–30 g/day, ≥30 g/day) and fruit/vegetable consumption (g).

Because exposure to silica may coexist with exposure to asbestos, we also conducted a sensitivity analysis testing whether observed associations could potentially be driven by asbestos exposure. For occupational groups exposed to silica, asbestos exposure was estimated using a procedure similar to that used for silica, using exposure information as given in table 1. For occupational groups not exposed to silica, that is, the reference group, it was not possible within the scope of the study to identify the occupations with asbestos exposure. For this reason, in the sensitivity analysis, non-exposure was defined as white collar work based on an adjusted EPG classification to also rule out asbestos exposure,20 21 instead of all other work without exposure to silica as used in the main analyses. Potential confounding by asbestos exposure was evaluated by using information on cumulative estimated asbestos exposure as a continuous variable, unweighted as well as weighted for probability of exposure. Effect estimates for silica exposure with and without adjustment for asbestos exposure were compared, using white collar workers as non-exposed workers in both cases.

Results

A description of the 1667 male lung cancer cases and 2034 male subcohort members is presented in table 2. On average, cases were older, were more likely to report a family history of lung cancer, were more likely to smoke cigarettes, smoked more cigarettes per day and for more years, had a lower consumption of fruit and vegetables, and were more likely to drink more than 30 g alcohol per day compared to members of the subcohort.

Table 2

Distribution of potential confounders and exposure among male members of the subcohort and of lung cancer cases

Overall, 210 cases and 228 subcohort members were exposed to silica during any one of their jobs. Out of the 27 occupational groups with assigned silica exposures in FINJEM, 21 were present in the NLCS population (see table 1). Most often reported jobs were ‘machine and engine mechanics’ and ‘construction carpenters’, followed by ‘bricklayers, plasterers and tile setters’. There was only a slight difference between the lung cancer cases and subcohort in average silica exposure concentration, but among cases the duration of exposure was on average longer. Hence, cumulative exposure was also higher among cases at 2.3 vs 1.6 mg/m3.year. Only about 10% of exposed people had a cumulative exposure of 4–5 mg/m3.year or higher. Within the restricted population (with ≥90% probability of exposure), duration of exposure was similar to that in the entire population, but average concentration and cumulative exposure were higher.

Overall, associations in the entire population between different silica exposure variables and lung cancer were positive for the full model including age, family history of lung cancer and lifestyle factors, but reached statistical significance only for exposure duration (table 3). Lifestyle factors acted in all analyses as a confounder; RRs were generally about 0.2 absolute points higher for categorical variables without adjustment for lifestyle factors. These changes could nearly completely be ascribed to the effect of adjustment for smoking. Additional adjustment for alcohol and fruit and vegetable intake had a small effect (data not shown).

Table 3

Lung cancer risk* related to silica exposure for entire population and restricted population based on exposure probability

For exposure variables on a continuous scale and for groups in the higher exposure range, except for mean exposure, associations approached statistical significance at the 0.05 level. Point estimates were slightly higher for the restricted population than for the entire population.

If both mean concentration and duration were simultaneously included in the model, the risk estimate for mean concentration was largely reduced as compared to the model with mean concentration alone. The risk estimate for duration was only slightly altered as compared to the model with duration alone. Associations for continuous exposure based on log transformed exposure variables (ie, concentration and cumulative exposure) did not differ markedly from those for non-transformed values (data not shown).

Additional analyses, based on the population with non-exposure limited to white collar work, gave no indication of confounding by asbestos exposure (table 4). Adjustment for asbestos exposure weighted for probability of exposure yielded even slightly higher risk estimates for the longest silica exposure duration (RR 2.28, 95% CI 1.06 to 4.87) and the highest cumulative exposure (RR 2.01, 95% CI 1.05 to 3.84) than without adjustment (RR 1.96, 95% CI 1.25 to 3.07 and RR 1.6, 95% CI 0.91 to 2.80, respectively). Risk estimates were slightly lower if asbestos exposure was not weighted for probability (data not shown).

Table 4

Effect of adjustment for asbestos exposure on association between lung cancer sand silica exposure

Discussion

Despite classification of silica as human carcinogen in Europe and the USA,22 23 and a large number of studies showing positive associations, there is still discussion on the actual carcinogenicity of crystalline silica.24 The role of silicosis in lung cancer aetiology, as well as the heterogeneity between studies in results including negative results, may partly depend on differences in the “inherent characteristics of silica or in external factors affecting its biologic activity”.23 Most positive associations have been found in occupational cohorts with rather high (cumulative) exposure and among patients with silicosis. Some population-based case–control studies also reported significant positive associations.9–11 25

To our knowledge this is the first prospective, population-based study in which the association between lung cancer and silica exposure has been studied and demonstrated. In our study, the majority of the specific associations with different exposure variables point in the same direction. Results indicate elevated risk in the exposure groups with longest exposure and highest cumulative exposure, and to a lesser extent with higher average exposure levels. Most associations are not statistically significant at the 0.05 level, which may partly be attributable to the relatively small exposed population.

Associations are observed in a population mostly exposed before the 1980s. Occupational exposure tends to decrease over time,20 which could mean that the risk of lung cancer among workers currently exposed to silica may be lower. However, especially for the construction sector, which employed the largest group of exposed workers in our study, recent studies suggest that exposures to crystalline silica are still high and do not follow generic trends of significant exposure reduction over time.26–28 For current exposure situations, similar exposure–response relationships may therefore be expected.

As opposed to many industrial cohort studies, a major drawback of population-based studies is the lack of actual exposure data. In our study exposure was assigned based on a limited number of occupational groups as defined in FINJEM. This JEM was originally designed to be used for the Finnish population, each group representing several jobs. Original job classification in the NLCS was based on self-reported job title, name and type of company and products made. This methodology may lead to several types of bias. One of the authors (LP) selected jobs out of a list of some 14 000 and linked them manually to the 70 Finnish occupational subgroups with assigned silica exposure, using rather detailed descriptions on Dutch job contents in case of uncertainty.29 It is not expected that this linkage led to much misclassification of exposure status (exposed/not exposed). Only Dutch jobs with exposure not included in FINJEM, for example sand and shingle extraction, might have been excluded incorrectly. Therefore, overall little bias is expected in associations with exposure status and duration of exposure.

The validity of FINJEM is optimised by using some 5000 exposure measurements on silica over the period 1972–2000 and the judgment of experts.15 Use of the Finnish data for the Dutch population may have led to bias in our study for average and cumulative exposure.

Results are presented as if exposures in our Dutch population were as indicated in FINJEM. Construction work is the main source of exposure in the restricted population. Measurements among Dutch construction workers in various jobs in recent periods indicate that exposure levels in the Netherlands are likely within the same range as in Finland for the same period. Reported full shift geometric mean exposure in a Dutch population was 0.13 mg/m3,30 and 0.5 mg/m3 for task-based exposure.31 Considerable bias in the estimate for average and cumulative exposure in the restricted population with a ≥90% probability of exposure seems therefore unlikely. However, the weaker associations for cumulative exposure as compared to exposure duration, may suggest that use of the external, quantitative JEM results in more (non-differential) misclassification, leading to underestimation of exposure–response relationships.32

Non-differential misclassification in individual exposure assessment is expected to result from assignment of exposure at a high aggregation level of occupational group in FINJEM. Using this level of aggregation, differences in exposure between jobs within a group and between workers within a job are disregarded.33 34 This type of error would lead to attenuation of exposure–response relationships.32 35 Attenuation is potentially larger than in case–control studies, since these may capture more specific information on actual occupational exposure by means of in-depth interviews specifically directed at occupational exposure; for example, Parks et al36 showed empirically that including more job and task information led to stronger associations between silica exposure and disease, although their health endpoint (systemic lupus erythematosus) depends less on chronic exposure than lung cancer.

Our prospective and population-based study has several major advantages. The prospective design rules out differential recall or interviewer bias in occupational history, potentially a source of bias in case–control studies. Error in occupational history will therefore most likely be non-differential, generally only leading to underestimation of exposure–response relationships.

Unlike most studies on industrial cohorts, we were able to adjust for smoking habits. Since smoking is such a strong determinant of lung cancer, and therefore potentially also a strong confounder, accurate data on smoking are needed to preclude major bias by smoking. As for occupational history, the quality of smoking information in our study does not depend on disease status. Adjustment for smoking reduced estimates of RRs by 0.2–0.3 absolute points. Other population-based studies showed similar effects of adjustment.10 11 Yu et al,13 however, observed that risk in smokers was overestimated, but that risk in non-smokers was underestimated. Other lifestyle factors recognised as risk factors for lung cancer (ie, fruit and vegetable consumption and alcohol intake) appeared not to be confounders within our population. Since virtually all studies indicate that silica is a rather weak carcinogen with risk estimates of 2 or lower for highly exposed workers, limiting these sources of bias is very important for detecting positive associations.

Co-exposure to other occupational carcinogens and the presence of silicosis have also been suggested as potential sources of bias. The presented associations of the main analyses do not seem to be inflated by co-exposure to asbestos. Sensitivity analyses on a subpopulation show even somewhat stronger risk estimates for silica exposure in the models with adjustment for asbestos exposure than in those without adjustment. Other population-based studies using JEMs for estimation of silica and asbestos exposure were in line with these findings and did not show strong confounding of silica exposure by asbestos exposure.10 11 We did not adjust for exposure to other occupational carcinogens and could not study the effect of the presence of silicosis. Bias by these items cannot be ruled out.

Associations were slightly to moderately stronger in the population restricted to ≥90% probability of exposure than in the full population. About 25% of cases and subcohort members were omitted from the full population in an attempt to reduce exposure misclassification. Especially when non-exposed subject are classified as exposed (when exposure is relatively rare), and this misclassification is non-differential, this leads to underestimation of the true exposure–response relationship.19

Exposure–response relationships found in our study are in line with many studies on lung cancer and silica exposure. Information on exposure–response relationships is mostly based on industrial cohorts. In some of these cohorts, exposure can be measured rather precisely because of the availability of measured exposure data and specific occupational history. Steenland et al8 pooled the data of 10 industrial cohorts. In the individual studies, risk estimates (SMRs) were often between 1 and 2, and the pooled data showed ORs of about 1.6 in the highest exposure categories compared to the lowest exposed group, its size depending on the exact exposure variable and analysis. High exposure in that study was, however, much higher than in our study. Forty per cent of that population was estimated to be exposed to ≥5.4 mg/m3.year, whereas in our study only about 5% of those exposed were exposed to levels >5 mg/m3.year.

Only a few population-based studies focused on the association between silica exposure and lung cancer, with risk estimates in the same range as in our study. The Finnish study, based on cancer registry and census information, reported an elevated risk of 1.4 in the group with highest cumulative exposure, of >10 mg/m3.year, a level which is high compared to our highly exposed workers.11 That study had the important advantages of having exposure estimates specifically based on Finnish working conditions and directly specified for the distinct occupational categories. They lacked, however, information on actual individual occupational history and smoking habits.

Despite more potential attenuation of exposure–response relationships by non-differential misclassification of exposure in our study, risk estimates for comparable exposure levels are in the same range as in a German case–control study,10 and somewhat lower than in a recent multicenter case–control study.9 The German study found an OR of 1.4 for workers in the medium exposed group (1–5 mg/m3.year) and an OR of 1.9 in the group exposed to a level of >5 mg/m3.year. The multicenter study, mostly based on eastern European regions, yielded the highest risk estimate of 2.1 in the highest quartile of exposed workers, but their exposure categorisation is not directly comparable to that in the other studies including ours. Among non-smokers in basically the same populations, the risk estimate in the highest exposure group was 2.4.25

In light of these results, our study, as far we know the first prospective population-based study on this subject, corroborates the results of recent studies indicating that silica should be considered a human carcinogen.

Acknowledgments

We thank Elleny Balder for her role in data analysis, Swenneke van den Heuvel for job coding and Erik Tielemans for his contributions in various stages of the study. We are indebted to the participants of this study and further wish to thank the cancer registries (IKA, IKL, IKMN, IKN, IKO, IKR, IKST, IKW, IKZ and VIKC), and PALGA. We also thank Dr A Volovics and Dr A Kester for statistical advice; Dr L Schouten, S van de Crommert, H Brants, J Nelissen, C de Zwart, M Moll, W van Dijk, M Jansen and A Pisters for assistance; and H van Montfort, T van Moergastel and R Schmeitz, for programming assistance.

References

Footnotes

  • Funding The Dutch Ministry of Social Affairs and Employment provided funding for this study. The NLCS was established with the financial support of the Dutch Cancer Society, Amsterdam.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of both the Netherlands Organisation for Applied Scientific Research TNO (Zeist, The Netherlands) and Maastricht University (Maastricht, The Netherlands).

  • Provenance and peer review Not commissioned; externally peer reviewed