World asbestos production declined as consequence of the increasing restrictions on its use from about 5 million tons/year in the 1980s to less than 2 million tons/year in 1999.1 So far, 30 countries worldwide have banned the use of asbestos, including Italy since 1992 and the European Union since 1991 (amphibole asbestos) and 1999 (all types). Therefore, a large number of workers had their exposure to asbestos terminated at a relatively young age and should be studied in order to provide information on the variation of cancer risk long after cessation of exposure, an issue that has been considered in a limited number of studies.2

Worldwide, the asbestos cement (AC) industry has been the biggest user of asbestos, with the greatest number of exposed workers. In Italy in the 1980s 120 000 tons/years of AC pipes and 700 000 tons/year of other AC goods were produced by a total workforce of about 5000 people.3 The present paper provides data on mortality by specific causes of death and on the incidence of mesothelioma among workers of the Eternit plant in Casale Monferrato (Piedmont, Italy), one of the most important AC plants in Italy.4 The Eternit factory ceased production in 1986, when a large section of the cohort was still of working age. As there were no other factories producing asbestos products in the area,5 the workers ceased to work in the industry and no longer experienced occupational asbestos exposure.6 This situation provides a unique opportunity to study the long-term effects of exposure after the cessation of exposure. In the present study we explore the role of duration of exposure, latency and time since the end of exposure in the causation of lung cancer and mesothelioma. We also study the effects of asbestos on cancer sites whose association with asbestos is still debated, such as the digestive tract,7 larynx8 and ovary.9

METHODS

The plant

The Eternit plant of Casale Monferrato was active from 1907 to 1986. It produced plain and corrugated sheets, chimney tubes and high-pressure pipes using chrysotile and crocidolite. All activities ceased in 1986. Data on airborne asbestos concentrations in the plant showed high exposure, but there were insufficient data to estimate individual cumulative doses.4 In 1971, the concentration of airborne asbestos fibres was found to be above 20 f/ml in 11 samples out of 22. Average concentration was 13.5 f/ml in the production areas and 303.8 f/ml in the area where asbestos and cement were dry mixed. In 1973, the averages of repeated measurements of asbestos fibre concentration were 13–15 f/ml in the mixing department, 1.2–1.8 f/ml in the production department and 0.7–1.1 f/ml in the finishing department. The company reported that working procedures were subsequently improved in order to reduce dust pollution. Regular monitoring of the concentration of airborne asbestos fibres by company staff started in 1978, when average total concentrations of asbestos fibres were reported in the range 0.15–1.12 f/ml in the mixing department, 0.18–1.05 f/ml in the production department and 0.29–1.09 f/ml in the finishing department. Results of measurements in the following years are reported to have not changed significantly.

Cohort description

The cohort included 3434 blue-collar workers (2657 men and 777 women) who were active at the plant on 1st January 1950 or were hired between 1950 and 1986. These workers were identified on the basis of the factory rosters, where all workers were recorded on the day of hiring and on the day of cessation. Only nine subjects listed in the factory rosters were excluded from the cohort because of lack of information. These rosters were also the source for defining job histories. Vital status and cause of death were ascertained through registrar offices. All subjects contributed person-years until their most recent date of observation. The underlying cause of death was coded by us according to the International Classification of Diseases (Ninth Revision).

Statistical analyses

Analyses are based on computation of the number of person-years and either standardised mortality ratios (SMR), standardised incidence ratios (SIR) or Poisson regression models.

For SMR analyses, the number of deaths expected in the cohort was estimated on the basis of mortality rates in Piedmont, provided by the National Institute of Statistics (ISTAT). Analyses were restricted to 1965–2003, as reference rates were available only for the period 1970–2002; for the period 1965–69 we applied the rates for the period 1970–1975. Rates for the period 2000–2002 were applied to 2003. SIRs were computed for malignant mesothelioma (MM) using the incidence rates provided by the Mesothelioma Registry of Piedmont.10 The 95% confidence intervals (95% CI) were estimated assuming the Poisson distribution of the number of observed deaths.11

Throughout the paper we used “latency” as “time since first exposure” and “duration of exposure” as “the sum of the individual working periods”. Duration of exposure therefore corresponds to the total duration of employment in the factory. “Year of first exposure” is defined as the calendar year when the worker was first hired to work in the factory. In most analyses we grouped this variable in periods (defined as “period of first exposure”).

Poisson regression analyses were based on person-years and numbers of events. Only cancer types clearly associated with asbestos exposure (ie, pleural and peritoneal malignancies and lung cancer) were considered. Analyses focused on the role of age, period, gender, duration of exposure, latency, period and age at first exposure, and time since last exposure. Reference categories were chosen a priori in order to avoid categories with few subjects. The contribution of a variable to the model fit was evaluated using the likelihood ratio test.11 Model building always started from a simple model including age (as continuous variable), gender and calendar period (as continuous variable). Further, we added in separate steps duration of exposure, latency and time since cessation of employment (one at a time). Later on, we combined two of these variables and, eventually, all three variables. Due to the possible collinearity among the variables tested in the multivariate regression analyses, in defining the best model we considered both the statistical significance of the contribution and also if the addition of a new variable unexpectedly changed the estimated coefficient and its standard error for the variables already in the model.

Analyses were carried out using OCMAP PLUS v 3.10 (University of Pittsburgh, PA, USA) and STATA 8 (Stata, College Station, TX, USA).

RESULTS

A total of 3434 subjects (2657 men and 777 women) were included in the cohort. Follow-up was completed for 98.7% of the subjects. The cause of death was known for 98% of deceased subjects. In contrast to other studies of asbestos workers, the workforce was very stable. Table 1 provides descriptive information on the cohort. During 1965–2003, the cohort contributed 63 998 person-years for men and 22 367 person-years for women.

Table 2 presents the SMRs for the main causes of death. The table includes causes selected on the basis of a priori evidence of association with asbestos, causes associated with the healthy worker effect (HWE)12 and causes that showed a statistically significant SMR in exploratory analyses. Both genders showed increased mortality for all causes (p<0.01), all malignancies (p<0.01), pleural and peritoneal malignancies (both p<0.01) and lung cancer (p<0.01 for men and p<0.05 for women). In women, ovarian and uterine malignancies were more frequent than expected (p<0.01 and p<0.05, respectively). Rectal cancers were more frequent than expected in women but not in men and the excess was not statistically significant after pooling data. No statistically significant increase was found for laryngeal cancer in either genders or after pooling. Asbestosis was the underlying cause of death for 162 men and 24 women (crude mortality rates: 253.1 and 107.3 per 100 000 person-years) and a contributory cause for 309 men and 52 women. The number of deaths due to cardiovascular diseases was lower than expected in both genders, which is indicative of HWE.

Forty nine cases of pleural MM were observed: 43 (82.7%) were diagnosed on the basis of histology and six (11.9%) on the basis of cytology (data not tabulated). Corresponding figures for the 23 peritoneal MM were 20 (87.0%) and three (13.0%). A further five cases (three pleural) diagnosed on a radiological basis were not considered. SIRs for pleural MM were 3447 in men (37 observed (obs) vs 1.07 expected (exp)) and 5952 in women (12 obs vs 0.20 exp); SIRs for peritoneal MM were 17 805 in men (18 obs vs 0.10 exp) and 18 592 in women (5 obs vs 0.03 exp).

Table 3 presents SMR analyses by latency for selected causes of death. A large proportion (20%) of the deaths in the cohort occurred after 50 years of latency. A statistically significant increase in SMRs in the longer periods of latency was observed in both genders for all causes, all malignant neoplasm, peritoneal and pleural neoplasm and asbestosis. SMRs for lung cancer showed a curvilinear trend with maximums at 20–29 and 30–39 years. A statistically significant increase in mortality for pleural neoplasm was observed beginning in the class 10–19 years of latency; the medians were 36 and 45 years, respectively, in men and women. Regarding peritoneal neoplasm, SMRs become statistically significant after 20 years of latency; the medians were 40 and 53, respectively. The number of deaths from “all causes” and “cardiovascular diseases” was lower than expected in the first 10 years of latency, suggesting a marked HWE.

The analyses by duration of exposure (table 4) in men showed a statistically significant increase in SMR after 1–4 years of exposure for pleural and peritoneal malignancies, and for asbestosis, while the increase in SMRs for lung cancer becomes statistically significant after 5–9 years of exposure. In women a statistically significant increase in mortality for these causes was also observed, with patterns similar to those observed in men. Twelve cases of pleural or peritoneal neoplasm in men and four in women occurred after less than 5 years of exposure.

Poisson regression analyses were carried out in order to study the joint effect of duration of exposure, latency, time since last exposure and other time related variables on lung, pleural and peritoneal neoplasm mortality (table 5). All subjects in the cohort were included and therefore absolute numbers can be larger than the figures considered in SMR analyses, which were limited to the period 1965–2003. In the interpretation of the relative risks (RRs) from these analyses, we should remember that they measure the variation of the risk within the cohort; as everybody experienced some asbestos exposure, these results cannot be taken as measures of the risk in reference to truly non-exposed subjects. Selected models included only the variables that were statistically significant in the model fit.

The RR for pleural neoplasm increased linearly according to duration of exposure (table 5A). The effect of latency was curvilinear, suggesting a reduction in RR after 50 years of latency, with wide confidence intervals. A trend towards reduced risks after over 15 years since the cessation of exposure was also observed but confidence intervals were wide and the RR was not statistically significant. As the effect of duration of exposure was linear, we also fit it as a continuous variable: the RR was 1.05 (95% CI 1.01 to 1.08) per year of exposure.

Regarding peritoneal neoplasm, a marked trend in RRs was observed according to duration of exposure and latency (table 5B). Period of first exposure and time since last exposure did not provide a significant contribution to the model. The trend by latency is much steeper than observed for pleural mesothelioma and there are no indications of lowering risks after long latency or time since last exposure (details not given).

Regarding lung cancer (table 5C), the selected model included time since last exposure, latency and gender. Latency showed a curvilinear trend, with the highest RR for the class 30–39 years. We also included in the model year of birth (in classes) in order to test for possible cohort effects due to changing smoking exposure: the result showed the expected cohort effect with lower RR in the more recent birth cohorts, but the contribution to the model was not statistically significant and the confidence intervals were wide (details not given). Also the model including time since last exposure, period of first exposure and gender was statistically significant but the effect of period of first exposure is likely to be confounded by latency (details not given).

DISCUSSION

This cohort study of the workforce of the Eternit plant has provided an important opportunity for investigating the risk of asbestos related diseases by latency, duration and time from cessation of exposure in a large and homogeneous population with long employment and follow-up. Compared to our last report,4 this study added 10 years of follow-up and new data on the incidence of MM as well as new statistical analyses. When operations ceased, no other firms using asbestos were active in the area and workers’ exposures were effectively terminated.5 6 13 We could not exclude the chance that individuals might have been exposed to asbestos in other jobs, but such an occurrence is likely to be similar to that observed for the general population.

The design and data source of this study make major biases unlikely. The cohort consisted of those recorded on the original factory rosters, with no selections. Out of 3443 subjects listed, 3434 (99.7%) were be included in the study. Follow-up was carried out using official sources that were not related to the factory and were not influenced by the subject’s inclusion in the cohort or health status. Vital status was ascertained for 98.7% of the cohort and the cause of death was known for 98.0% of the deceased subjects. Mortality in the cohort was compared to regional rates, which are more appropriate because of the wide regional differences in respiratory cancer mortality in Italy.14 As regards pleural cancer, comparison with the regional rather than the national population is also more appropriate because mortality from pleural cancer is higher in Piedmont, and in general varies widely among Italian regions.15 Piedmont has a population of 4.5 million and mortality rates are not affected by mortality in the cohort under investigation. Casale Monferrato is a small town (40 000 inhabitants) separated from the main industrial areas of Northern Italy.5

Regarding MM incidence, both incident cases and incidence rates were ascertained from the regional mesothelioma registry. SMRs in our study did not over-estimate RR when compared to SIRs (details not given) and therefore the main analyses were based on mortality data because of their longer availability and greater statistical stability due to larger numbers.

Our analyses were based on duration of exposure (defined as duration of employment in the factory), which is a proxy of cumulative exposure. In studies encompassing such a long period of observation, the exposure estimation given by duration is often the only choice and may not be improved by exposure measurements because of their imprecision.11

As the factory used both chrysotile and crocidolite, which were mixed in the proportion required for the different final products, in the present study the carcinogenic roles of chrysotile and crocidolite cannot be disentangled.

Total mortality SMRs were 135.0 in men and 149.5 in women; these figures corresponded to almost 480 extra deaths among the subjects in the cohort over a period of 41 years. The excess mortality was due to an increase in mortality from pulmonary, pleural and peritoneal malignancies and asbestosis. Analyses of both total and cardiovascular mortality indicated a substantial HWE, observed independently of age at first exposure and period of hiring (data not shown), which was stronger in men than in women.

Our SMRs for lung cancer (242.5 among men and 220.5 among women) are close to the results of other cohort studies of AC workers exposed to a considerable amount of amphiboles.16^–25 A meta-analysis8 estimated a pooled SMR of 158 (95% CI 144 to 173) for lung cancer in AC workers.

SMRs for lung cancer by exposure and latency showed a curvilinear trend, which decreased after 40 years of latency and 30 years of exposure. Poisson regression provided corresponding results showing a statistically significant association with gender, time since the cessation of exposure and latency. The eightfold increase in RR for lung cancer in men compared to women is likely to reflect the effect of smoking, which was more frequent in men than in women.

A reduction in the SMRs as was observed for lung cancer after 30 years of exposure, can have different explanations. The reduction could be due to the fact that the subjects employed in the factory for a longer time may have been given duties associated with lower exposures to asbestos. The decrease could also be due to the attrition of smoking related diseases: the combined exposure to asbestos and cigarettes could be so lethal that heavy smokers were eliminated from the study cohort, leading to the selection of non-smoking workers over the years of employment. This differential attrition could contribute to the observed decrease in the excess relative risk of lung cancer in the last categories of duration of exposure and latency. Duration of exposure did not reach statistical significance in the Poisson regression model, but when it was retained in the model (data not shown), the RRs for classes of duration showed a similar trend as in the corresponding SMR analysis (table 4).

The reduction in risk for lung cancer with increasing time since the end of exposure is relevant regarding the possible mechanism of action. In evaluating its impact, however, the reader should consider that Poisson regression was based on internal comparison and did not correspond to a reduction in the excess mortality to the same level as in the general population: among men the SMR after 30 years since the end of exposure was still 217.8 (21 obs vs 9.6 exp, p<0.01). The lack of local mortality rates for the general population before 1970 (see Statistical analyses section for details) precluded the use of the SMR as offset in Poisson regression analyses. Few studies reported comparable observations on risk according to time since the end of exposure. Pira et al2 observed lower SMRs in a cohort of asbestos textile workers after 15–25 years since the cessation of employment. In a case-control study, Hauptmann et al26 observed that the highest risk for lung cancer was related to exposure that occurred 10–15 years previously.

We observed the highest SMR for lung cancer after 30–39 years of latency in men and after 20–29 years in women, followed by a reduction. Corresponding results were observed in multivariate Poisson regression. Other studies had observed a decline in RRs with very long latency,2 22 24 27 28 however our study is one of the very few presenting results for over 50 years of latency. Walker27 tentatively interpreted the decline as a reduction in lung cancer risk at some time after asbestos exposure cessation.

Both mortality and incidence of pleural and peritoneal MM are increased in both genders in this cohort. Pleural and peritoneal neoplasm caused 6.7% and 2.5%, respectively, of the total number of deaths in men and 10.5% and 4.3% in women. Other studies of AC workers with substantial exposure to amphiboles reported similar figures.8 29 In our study, Poisson regression analyses showed different patterns for pleural and peritoneal neoplasm. RR for peritoneal neoplasm showed a linear increase with duration of exposure and latency. RR for pleural neoplasm was also associated with duration of exposure and latency, but the relationship with latency was curvilinear. The different patterns of pleural and peritoneal neoplasm is a clear indication against pooling the two diseases in the analyses.

Some of the variables included in the Poisson regression model are associated with each other. In order to deal with the possible collinearity among the variables, we always considered if the addition of a new variable unexpectedly changed the estimated coefficient and its standard error for the variables already in the model.

Berry suggested the possibility of a decline in risk of MM after a long time since exposure due to the elimination of asbestos from the lung.30 A recent update of the Wittenoom cohort provided some support for this elimination model,31 but results were not statistically significantly because of the small numbers. Our results for pleural neoplasm also suggest a decline in risk after long latency; the suggestion will be further explored in future investigations. No such trend was suggested for peritoneal neoplasm.

The present study did not show an association with laryngeal cancer. Our result was in agreement with another study on AC workers in Italy25 which observed an SMR of 72 (95% CI 15 to 210), and with a recent study on crocidolite workers which did not show an association after controlling for smoking habits.7 Conflicting results have been reported: a meta-analysis estimated a pooled SMR of 157 (95% CI 95 to 245) in asbestos workers, increasing with latency over 10 years.8 An international case control study showed an OR of 1.8 (95% CI 1.2 to 2.5) for asbestos exposure, but the OR decreased when analyses were limited to over 10 years of exposure and 20 years of latency (OR 1.4, 95% CI 0.8 to 2.4).32 Both laryngeal cancer and lung cancer are causally associated with smoking and therefore the potential for confounding exists. No individual information on smoking habits was available for the study. Nevertheless, there are no indications that the proportion of smokers in the factory was lower than in the general population and, moreover, if any confounding were present it should act in the same direction for both laryngeal and lung cancer.

In women we observed a significant increase of deaths from malignant neoplasm of the uterus and of the ovary. The SMR for ovarian neoplasm was >2, those neoplasm arising in women with at least 20 years of latency. The diagnosis was histologically confirmed in 7/9 cases. Albeit limited, evidence in the literature supports the causality of the observation: Heller et al33 documented that asbestos fibres reach the ovary and an excess of ovarian neoplasm among asbestos exposed women has been observed by other authors, including Germani et al,34 Berry et al35 and Pira et al.2 Interpretation of the excess of uterine neoplasm is uncertain. The cases arose after at least 10 years of latency, but the trend with duration of activity is unstable. No clinical record was found for revision of diagnosis in 5/15 cases. Previous evidence in the literature is very limited: three studies23 28 36 reported an increase in mortality from cervical cancer in asbestos exposed women. We could not find any published results on endometrial cancer and asbestos exposure.

In conclusion, the present study confirmed the previous observations of a large excess of lung cancer, pleural and peritoneal mesothelioma and asbestosis in the cohort under investigation. It also ruled out a possible association with laryngeal cancer or cancer of the digestive tract. The long follow-up allowed some new analyses of time related variables that indicate a reduction in risk for lung cancer after the cessation of exposure and confirm the role of duration of exposure in the aetiology of pleural mesothelioma. Pleural and peritoneal mesothelioma show distinct patterns in risk after long latency (increasing for peritoneal MM but not for pleural) which may be obscured if the two diseases are pooled.