Occupational exposure is considered to be an important cause of lung cancer. Whereas smoking is considered to cause 85–90% of male lung cancer, various studies suggest that occupational exposure is responsible for 5–20% of cases.1^–5

In the past decades, several lung carcinogenic compounds have been identified, such as asbestos, chromium VI, beryllium, nickel, cadmium and silica. This information is generally derived from epidemiological studies in large industrial cohorts, and can serve to reduce exposure to such substances with the goal of reducing or eliminating occupational (lung) cancer. Risks attributed to occupational exposure to carcinogenic substances do not provide us with an estimate of the overall burden of lung cancer of occupational origin. Policy makers increasingly require this type of information and want to know how this burden is distributed across population subgroups. For example, disease burden estimates provide an indication of health gains that could be achieved by targeting action against specific risk factors.6

A few years ago occupation related mortality in the Netherlands was estimated in two studies. By extrapolating information from other countries, one group of researchers estimated that less than 9007 and the other that about 2000 annual lung cancer deaths8 could be attributed to occupational exposures. Since it is difficult to estimate the number of people occupationally exposed to lung carcinogens at a national level, information on the burden of lung cancer can be derived from data on risks in industrial sectors. Variation in the outcomes of studies in different countries on lung cancer related to specific industries9 10 indicates that these results cannot be extrapolated to the Netherlands. In addition, it is often not clear to what extent other strong determinants of lung cancer risk, such as smoking, contribute to this variation.

The aim of the present study was to estimate male occupational lung cancer risks among industrial sectors in the Netherlands compared to the working population as a whole. Also, we wanted to estimate the proportion of lung cancer attributable to working in sectors with elevated risk. We used data from the Netherlands Cohort Study (NLCS), a prospective cohort study on cancer and diet.11 The NLCS database contains detailed individual information on occupational history, smoking habits and other potential confounders. To obtain the best possible risk estimates we adjusted associations for different sets of potential confounders.

METHODS

Study population

The NLCS was approved by the institutional review boards of the Netherlands Organisation for Applied Scientific Research TNO (Zeist) and Maastricht University (Maastricht). 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.12 This group represented about 36% of the people invited to participate. Because the proportion of long-term employed women and the number of female lung cancer cases was quite small, the analyses are restricted to male lung cancer. In total 58 279 men completed a self-administered mailed questionnaire on their diet and potential risk factors for cancer, such as smoking, lifestyle characteristics, education and occupational history. This cohort study was designed as a case-cohort, in which the person-years at risk accumulated in the cohort are estimated from a subcohort, as decribed elswhere.11 After the baseline exposure measurement, a subcohort of 5000 subjects (2411 men) was randomly sampled from the large cohort. This subcohort was followed for migration and vital status to calculate person-time at risk. Two male subcohort members were lost to follow-up during the study period.

The entire cohort has been followed up for incidence of cancer by a method described previously.13 In short, incident cancer rates have been established by record linkage to the Netherlands Cancer Registry and the nationwide network and registry of histo- and cytopathology in the Netherlands (PALGA). After 11.3 years of follow-up (ie, from September 1986 to December 1997), 1920 incident primary male lung carcinoma cases (ICD-O-3 code: C34) were available for analysis. Prevalent cancer cases other than non-melanoma skin cancer were excluded, leaving 2251 men from the subcohort for analysis.

Assessment of job history

Occupational history was assessed by questions on the name and type of company, products produced in the department, job title, and period of employment. For each person, information on a maximum of five occupations was registered. In the few cases where more than five occupations were mentioned, similar consecutive jobs for different employers were combined and apparent sideline jobs were deleted. If more than five jobs still remained, the job with the least information provided was omitted unless it lasted for a very long time.

The type of industry was coded according to the Standard Industrial Classification (1974) and occupation was coded according to the Standard Occupational Classification (1984), both from the Dutch Central Bureau of Statistics (CBS).

For statistical analyses, all reported occupations were aggregated into 26 (two-digit) industrial categories. To help identify jobs with the highest probability of exposure, all occupations were defined as blue or white collar jobs by means of an adjusted EGP classification.14 15

Statistical analysis

Cox proportional hazards models were constructed to estimate rate ratios (RRs) and 95% confidence intervals (95% CIs) relating industrial sector to the incidence of lung cancer (STCOX procedure, Stata version SE 8.0; Stata, College Station, TX). The 95% confidence intervals were corrected for the additional variance introduced by using person-time estimates from a randomly sampled subcohort instead of the complete cohort,11 by means of the robust option.

RRs were estimated for total duration of employment in specific occupational groups.

For each of the 26 industrial categories, RRs were estimated for the blue collar workers compared to all others, that is, those who never worked in that type of industry and those who had worked in that type of industry but only as a white collar worker.

All associations were adjusted for age (years) and family history of lung cancer (yes/no). Smoking habits (current cigarette smoking at baseline (yes/no), number of cigarettes smoked per day (n) and years of smoking cigarettes (years) simultaneously in the models), consumption of fruit and vegetables (g/day) and alcohol consumption (0–30 g/day, >30 g/day) were additionally considered as potential confounders.

The effect of adjusting for the highest attained level of education, and the average physical activity level of all jobs held (least, medium and most active, based on a coding system by Hettinger et al16) was also tested. In addition, a reference group of “all other blue collar workers” instead of “all others” was used to investigate potential residual confounding.

To prevent loss of observations, dummy variables were constructed for missing covariable values. The number of missing values can be inferred from table 1, and ranged for different variables from 0 to 15% for years of smoking among subcohort members.

Risk estimates (and corresponding confidence limits), in combination with prevalence of exposure, were used to estimate the fraction of lung cancer incidence attributable to occupational exposure (attributable fraction, AF), using the formula:

where p is the fraction of exposed subjects in the subcohort.17

RESULTS

The 1920 male lung cancer cases and 2251 male subcohort members are described in table 1. On average, those with lung cancer 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, were more likely to drink more than 30 g alcohol per day, had a lower educational level and were on average more physically active within their occupation compared to members of the subcohort.

The RRs of lung cancer and their 95% confidence intervals, according to duration of employment in blue collar jobs in 26 specific industries, are presented in table 2. After adjustment for potential confounders (full model), 15 years or longer of blue collar work was associated with an increased risk of lung cancer in a number of industries. Industries that showed a statistically significant increased risk were “electronics and optical instruments industry”, “construction and homebuilding business”, and “railway company”. Industries for which a non-significant, but considerably increased risk (50% or more) was observed were the “mining, quarrying, offshore industry”, the “building materials, glass, clay, stone industry” and the “shipbuilding, motor vehicles, aircraft and transport industry”.

Statistically significant decreased risks were observed for blue collar workers in the “post office and telecommunication industry” and the “textiles and leather industry”.

Adjustment for confounders affected risk estimates most strongly (>0.3 absolute change) for the construction (decrease), electronics (increase) and railway sectors (increase). By far the largest change was caused by adjustment for smoking; models with age, family history of lung cancer and individual smoking data yielded risk estimates of 1.63 (95% CI 1.20 to 2.20), 2.02 (95% CI 1.21 to 3.38) and 2.26 (95% CI 0.95 to 5.40) for the construction, electronics and railway sectors, respectively. Adjustment for other lifestyle factors only marginally affected the associations. This pattern was similar for all other sectors where RR changed after adjustment for confounders.

In general, additional adjustment for education and the average physical activity level of jobs ever held affected the risk estimates only slightly and these variables were not included in the models. An analysis comparing blue collar workers in specific industries with all other blue collar workers only, did not show substantially different results either (data not shown).

An analysis in which the three sectors associated with an increased risk were included in the model simultaneously showed very similar risk estimates (data not shown). These results show that there is no confounding across sectors, that is, there is no or very little overlap between the sectors.

Attributable fraction

Using the RRs for working for ⩾15 years (full model) in the “construction and homebuilding business”, the “electronics and optical instruments industry” and the “railway company”, and the corresponding fraction of exposed subjects in the subcohort, the attributable fractions are 2.8% (95% CI 1.0 to 5.3), 1.5% (95% CI 0.3 to 3.3) and 0.7% (95% CI 0.0 to 2.7), respectively.

DISCUSSION

In our prospective cohort study, we found statistically significant elevated risks for blue collar workers with an occupational history of ⩾15 years in the “construction and homebuilding business”, the “electronics and optical instruments industry” and the “railway company”. High risks were also seen for the “mining, quarrying and offshore sector”, the “building materials, glass, clay, and stone production industry” and the “shipbuilding, motor vehicles, aircraft and transport industry”, but these were not statistically significant. A decreased risk was observed for the “post office and telecommunication sector” and the “textile and leather industry”.

Adjustment for lifestyle factors had a relatively large impact on some risk estimates. Although for most sectors the width of the confidence interval was barely affected by adjustment for lifestyle factors, for some sectors there was a shift in statistical significance from either non-significant to significant or vice versa. Adjustment for smoking had by far the largest effect on estimates; additional adjustment for fruit and vegetables, and alcohol consumption, education and physical activity during work had only minor effects.

The estimated attributable fraction was 5% for working 15 years or more in any of the three sectors with statistically increased risk.

Taking into account the period of data collection and the age of the participants, this study primarily describes associations with working circumstances from the 1930s to the 1980s.

Industrial sectors

An increased risk was observed for lengthy employment in the construction industry. This is in line with the results from some studies,3 5 9 although a decreased risk has also been observed.10 Some of the observed differences between studies can be caused by differences in exact industry definition. There are high numbers of bricklayers, carpenters and painters in our group of construction workers. Bricklayers are exposed to silica, an IARC class 1 lung carcinogen. Painting is also considered to be related to lung cancer, but the carcinogen(s) has not yet been identified.18 In a previous paper based on our NLCS population, with fewer cases, a positive association was suggested for high exposure to paint dust as opposed to painting in general.19 Little information is available on the risk for the specific group of construction carpenters. Wang et al20 assessed an elevated proportionate mortality ratio for construction carpenters, while Bruske-Hohlfeld9 and Firth et al21 reported an elevated incidence for the group carpenters and bricklayers together. Wood dust, one of the potential exposures of carpenters, generally is only found to be associated with cancer of the nasal cavities, but a recent study found a strong association with lung cancer.22

The observed high risk among workers in the electronics and optical instruments industry is unexpected. Among the NLCS population, this group encompasses a number of jobs, such as those of bench workers/metalworkers, electrical fitters, maintenance and related electronics workers, and makers of electrical or mechanical instruments. It cannot be ruled out that this is a chance finding. We therefore also tested associations for working for <5 years or for ⩾5 years in this sector, and observed for both groups a (non-significant) increased risk of 1.14 and 1.38, respectively. This, and the strength of the association independently of the used set of confounders, makes a chance finding less likely.

The relatively small group of 39 railway workers (both cases and subcohort members) consists mostly of railway equipment mechanics and operators. Asbestos and diesel have been suggested as potential causative agents of lung cancer in these groups,23^–26 but exposure to diesel exhaust is likely to be of minor relevance because of the early and widespread electrification of the railroad network in the Netherlands. A chance finding cannot be ruled out, but all presented RRs, including those unadjusted for smoking, suggest a positive association.

The decreased risk among workers in the textile and leather industry cannot be explained, although the suggested protective effect of bacterial endotoxins might play a role.27 The outdoor work with little exposure to carcinogens carried out by post office and telecommunication workers, whether or not in combination with high physical activity, might contribute to the low risk observed in this group.

The non-significant RRs of over 1.5 found in the “mining, quarrying and offshore sector”, “building materials, glass, clay, and stone production industry” and “shipbuilding, motor vehicles, aircraft and transport industry” might reflect a causal association. Elevated risks were also found in other studies5 9 10 28 29 and might be related to exposure to silica or asbestos.

Attributable fraction

Because there was very little overlap between the groups working for ⩾15 years in different sectors, we summed the separate AFs per sector to obtain an indication for the proportion of lung cancer cases in the cohort attributed to work related factors. Since we did not study exposure to specific agents and exposure varies within sectors, we cannot make inferences on causative factors. However, exposure to hazardous substances is likely to be one of these work related factors.

The overall AF is 5% for the three sectors with statistically significant elevated RRs. This is a rather conservative estimate because information on groups working for <15 years in a sector and on groups with a non-statistically significant elevated risk were not taken into account. Besides, since for each of the industrial categories RRs were estimated for the blue collar workers compared to all others, industrial sectors with elevated risks are included in the reference groups. Although the information on AFs cannot be used as exact estimates, this type of information can help policy makers understand the relative importance of the findings.

If we assume that the NLCS cohort is representative of the Dutch population with respect to blue collar work in and distribution across the sectors, we can extrapolate the AF from the NLCS cohort to the Dutch population of the same age and for the same period of time. For men aged 55–69 in 1986, there were about 40 000 lung cancer cases between 1986 and 1997, which is approximately 50% of the total Dutch male lung cancer incidence during that period.30 This implies that, within this period and in men of the same age range as the cohort, about 2000 lung cancer cases could be attributed to having worked in these sectors for 15 years or more.

Strengths and limitations

Our study has several advantages: the prospective design, the availability of individual data on potential confounders such as smoking habits and diet, and the population-based character for estimating the burden of disease. As far the authors know, this is the only population based study on occupational lung cancer risk which combines a prospective design with individual data on potential confounders. The prospective design rules out the effect of differential misclassification of exposure and potential confounders, which may lead either to under- or overestimation of associations.

Adjustment for dietary intake of fruit and vegetables and alcohol consumption barely influenced our results and RRs (and confidence intervals) changed only marginally. In general, cancer is found more frequently among people with a low socioeconomic status (SES) score. We tested the potential confounding effect of education as a proxy for SES score. We also calculated RRs with all other blue collar workers as the reference group instead of “all others” including white collar workers. Risk estimates were only slightly affected, supporting the conclusion that SES related factors other than those included in the analyses, do not explain the association between sector and lung cancer. Focusing on the two sectors with the strongest positive associations, it is obvious that adjustment for smoking can have very different effects, ranging from a decrease in RR from 2.0 to 1.6 for construction workers (with hardly any effect on the size of the confidence interval), to an increase from 1.6 to 2.0 in the electronics and optical instruments industry (with an increase in size of the confidence interval). Similar shifts, of generally <25% in estimates for RRs for lung cancer in sectors and jobs, have empirically been shown in case-control studies using individual data on smoking.9 31 32 Although shifts in RR and statistical significance may be relatively small, the effect on estimated burden of disease can be substantial when a large sector is considered. For example, if we had used the unadjusted RR of 1.99 in the construction industry, this would have resulted in an estimated AF of 4.3% instead of 2.8% for this sector, an overestimation of more than 50%. On the other hand, other lifestyle factors that are independently associated with lung cancer risk appeared to be much less important confounders in our study.

Baseline information was used to code occupational information and data on potential confounders. During study follow-up changes may have occurred in confounders, which might affect lung cancer risk. However, changes in occupation and exposure will be small due to the age of the cohort. Because of the prospective design, bias by errors and changes is likely to be non-differential leading mostly to attenuation of the risks estimated for the sectors. Incident cases are detected by record linkage to two national disease registries. Cancer registration is based on primary cancer cause, independent of presence in the NLCS population, and the registration rate is very high, making bias by the registration process unlikely. Selection bias can hardly be an issue because the follow-up of subjects is almost complete.

Since the AF depends on RR as well as on the proportion of exposed workers, bias might occur through errors in both of these parameters. The population originated from random samples from 204 municipalities throughout the country. Although the response rate was relatively low (36%), the distribution of demographic variables, smoking and dietary habits did not differ between the cohort and the general population.11 A small overestimation of the AF might have occurred through an over-representation of people with less education (generally more associated with exposure to hazardous substances) than of people with more education.19 This effect is probably low compared to the effect of the choice of model and corresponding RR to estimate the AF. In addition, the source population is derived from 204 of the 700 or so municipalities then existing in the Netherlands, but has a good geographical spread. Since working in construction is distributed rather evenly over the municipalities, it is not expected that this sampling procedure particularly affects the estimated AF. For the electronics sector, the potential inclusion of workers from a large company might lead to overestimation of the AF for this sector. On the other hand, exclusion of a region with one of the largest railway workshops might lead to underestimation of the AF for the railway sector.

In conclusion, this study shows increased lung cancer risks in several industrial sectors in the Netherlands. Taking into account the limitations of these data, our analyses suggest that about 2000 lung cancer cases which occurred in the Netherlands between 1986 and 1997 in the same age group as the NLCS, can be attributed to working for ⩾15 years in the three sectors with statistically significantly increased risk. Research on specific substances is needed to further investigate causative factors. In particular, the unexpected high risk in the electronics and optical instruments industry may give rise to additional research. Adjustment for individual smoking habits, but not for other lifestyle factors, seems important in reducing bias in risk estimates.