The importance of time-related factors in the development of asbestos-related diseases has been recognized for decades. Both cumulative exposure and time since first exposure (TSFE) to asbestos are related to the rate of mesothelioma. On the basis of the multistage model of tumor induction, Newhouse and Berry1 suggested that the risk for mesothelioma increases as a power of TSFE, according to:
where I is the incidence rate, c is the cumulative exposure, t is TSFE, w is the lag time, the interval after the start of the exposure during which it is assumed that the incidence of mesothelioma does not increase (usually 5 or 10 years), and k is estimated to be around 3.2 This model (hereafter referred to as the traditional model) predicts that the rate of mesothelioma starts to increase 5–10 years after the beginning of exposure and continues to increase subsequently, even after termination of exposure. Even though the findings of several studies provided support for this model,3 during the 1980s some authors4 suggested that the increasing trend in the incidence of mesothelioma with TSFE was not monotonic and that incidence rates could eventually start to decline many years after first exposure. In a modified model proposed by Berry,5 the increase in mesothelioma rate with TSFE was attenuated by a factor that represents the clearance of asbestos fibers from the lung. In this “elimination” model,
the elimination of fibers was taken into account by assuming first-order kinetics, in which the elimination rate λ was constant.
In order to test Berry's hypothesis, we used data from a cohort of workers formerly employed at the Eternit plant in Casale Monferrato (northern Italy), one of the largest European plants specialized in the production of asbestos-cement. The plant was active from 1907 to 1986 and provides a unique opportunity for studying the long-term patterns of mesothelioma incidence after exposure to asbestos. We compared the observed mortality rates from pleural and peritoneal cancer with the rates expected in the elimination model, in order to evaluate whether it fitted the data better than the traditional model.
The Eternit asbestos cement plant of Casale Monferrato produced plain and corrugated sheets, chimney tubes and high-pressure pipes. The raw material included both chrysotile and crocidolite, but not amosite. There have been reports that, during the 1970s, the consumption of crocidolite decreased rapidly and was used mainly for the production of high-pressure pipes; in the 1980s, according to company data, it accounted for only 10% of the total amount of asbestos used.6 Measurements made in 1971 showed that the airborne asbestos concentration exceeded 20 fibers per milliliter (ff/ml) in most production areas and was about 200 ff/ml in the area where asbestos and cement were mixed in a dry process.6 To reduce fiber concentrations, some changes were introduced into production processes from 1973, but regular monitoring procedures only from 1978.6
Cohort assembly and followup
The cohort comprised 3,443 blue-collar workers (2,663 men and 780 women) who were active on January 1, 1950 or hired between 1950 and 1986. Dates of birth and dates of start and cessation of employment were obtained from factory personnel records. Of the 3,443 persons listed, 3,434 (99.7%) were included in the follow-up. Each member of the cohort contributed to the person-years at risk since hiring (or January 1, 1950 if hired previously) until either death, migration abroad, loss to follow-up or end of follow-up (April 30, 2003). During 1950–2003, the cohort contributed 116,033 person-years (84,665 person-years for men and 31,368 for women), of which 62% were accumulated after cessation of employment in the plant.
Local registry offices provided the vital status and cause of death of each member of the cohort. Vital status was ascertained for 99% of the cohort, and the cause of death was known for 98% of the deceased subjects. At the end of follow-up, 1,960 subjects (57%) were dead, 1,441 (42%) alive, 7 (0.2%) lost to follow-up and 26 (0.8%) had moved abroad. The underlying cause of death was coded according to the International Classification of Diseases (9th revision).
Various Poisson regression models were fitted to the data to estimate mortality rates from pleural (ICD-9 163) and peritoneal cancer (ICD-9 158). We used 2 approaches to model the effect of TSFE. First, we conducted an exploratory analysis using TSFE as a categorical variable. The variables sex, age (5-year classes), duration of employment (1-year classes) and calendar year (5-year classes) were included in the categorical models. Calendar year was included in these models in order to take into account improvements in the accuracy of diagnosis of pleural and peritoneal cancer over time. Second, the elimination model was compared to the traditional model, which does not include a parameter for the elimination of asbestos from the lung. We set w equal to 5 years in both the traditional and the elimination model, as suggested by Berry et al.7 The parameters k and λ were estimated from the data. The likelihood ratio test was used to compare the elimination model containing 3 parameters (k, λ and the intercept) with the traditional model containing 2 parameters (k and the intercept), essentially testing λ = 0. In all analyses, time since start of employment at the plant was used as a proxy of TSFE to asbestos. Duration of employment at the Eternit plant (computed by summing all separate intervals) was used as a proxy for cumulative exposure. Statistical analyses were conducted with STATA version 8.0.
Table I shows the results of fitting categorical models. The rate ratio (RR) for pleural cancer showed a stead increase in the first 4 categories of TSFE and reached a plateau 40 years after first exposure. In contrast, the RR for peritoneal cancer increased monotonically over all categories of TSFE.
Table I. Poisson Regression Models: Rate Ratios and 95% Confidence Intervals (95% CI) for Pleural and Peritoneal Cancer by Time Since First Exposure (TSFE). TSFE is Modeled as a Categorical Variable
Time since first exposure (years)
Rate ratio (95% CI)
Rate ratio (95% CI)
Results adjusted for age, sex, calendar year and duration of exposure.
Table II shows the results obtained with the elimination and the traditional models. The likelihood ratio test suggests that the elimination model fitted the data better than the traditional model (p = 0.02) for pleural cancer. The maximum likelihood estimate of k in the elimination model was 2.95 for pleural cancer, in good agreement with the figures reported in other studies.3, 8 The maximum likelihood estimate of k in the traditional model for pleural cancer (k = 1.27) was, however, below the values reported in the literature, suggesting that the elimination model is more realistic and more biologically plausible. The elimination model estimated an asbestos elimination rate of 6% per year, corresponding to a half-life of asbestos fibers in the lung of 11 years.
Table II. Comparison of Elimination and Traditional Model Estimated Parameter Values and Their 95% Confidence Intervals (95% CI) for Pleural and Peritoneal Cancer
Estimated value (95% CI)
Comparison of two models (likelihood ratio test)
p = 0.02
p = 0.22
For peritoneal cancer, the elimination model did not fit the data better than the traditional model (likelihood ratio test p = 0.22). Moreover, the estimated parameter values in the elimination model (k = 0.74; λ = −0.05) were not biologically plausible, as a negative value for clearance rate is not interpretable and the estimated value of k is very low. Thus, in this case, the traditional model appeared to be more suitable. Figure 1 shows the observed rates of pleural cancer and the rates expected according to the elimination and traditional models against TSFE. When the traditional model was used to fit the data up to 40 years of TSFE (dotted line), it predicted a steep increase with TSFE, and the fit was excellent, whereas the observed rates after 40 years of TSFE were far below the expected values. When the observations after 40 years of TSFE were included in the traditional model (solid line), the curve obtained was almost a straight line (i.e. a value of k close to 1), which is not consistent with the results of similar studies and does not appear realistic from a biological point of view.9 The elimination model (dashed line) showed results similar to those of the traditional model up to 40 years but diverged thereafter, clearly reflecting the plateau in rates associated with the last categories of TSFE.
In this study, the rates of pleural cancer increased steeply up to 40 years after the beginning of exposure but remained fairly steady thereafter. This finding contradicts the common view that the risk for mesothelioma increases indefinitely with TSFE. There is a variety of possible explanations. Pleural cancers in old age, which were more common in the last categories of TSFE, might be poorly ascertained10; however, as age at death was included in the categorical models, the results are unlikely to be biased. Indeed, after age-adjustment, no increase in the RR for deaths from “unspecified causes,” “malignant neoplasms of unspecified site” or “lung cancer” were observed in the last categories of TSFE.
Several authors have pointed out that the rates of various types of cancer decrease in old age.11–13 The reasons for this phenomenon, if it is real, are still a matter of debate, but there is agreement that models based on the multi-stage theory fail to predict cancer rates in the elderly efficiently.13 Registry-based studies in various industrialized countries show that mesothelioma rates peak at 80 years and decrease sharply thereafter.14, 15 In the present study, however, after exclusion from the analysis of subjects aged more than 75 years, the estimate of parameters changed only trivially (in the alternative model, k = 3.2 and λ = 0.07).
Diagnostic awareness of mesothelioma may have increased over the years; however, most of the mesotheliomas among persons with more than 40 years of TSFE were observed during the last decade of follow-up, when diagnostic accuracy was high, the relation between asbestos and mesothelioma was well known, and claims for compensation for asbestos-related diseases were already common in Italy. All these factors would make it more reasonable to expect an increase in rates during the last period (and, hence, in the last categories of TSFE), rather than a decrease. Furthermore, when we limited the analysis to the period 1990–2003, in order to reduce possible misclassification of the diagnosis, the estimates of the parameters did not change substantially (for the alternative model, k = 2.4 and λ = 0.07).
It is unlikely that an improvement in survival could account for these results, as the prognosis after a diagnosis of mesothelioma has not improved perceptibly over the past 20 years.16
It could be questioned whether duration of exposure is a good proxy for cumulative exposure. In our study, almost all the person-years of observation in the last categories of TSFE were contributed by persons who started work at the plant before 1950, when exposure levels were higher than in subsequent years. For this reason, use of duration of the exposure as a proxy for cumulative exposure would be likely to underestimate the risks for persons with longer TSFE. This scenario would therefore explain an increase of mortality after a long TSFE and not the observed decrease. Furthermore, changes in the type of asbestos used in the Eternit plant in different periods of activity cannot explain the observed trends. Crocidolite, the most carcinogenic asbestos type, was used extensively in the plant until the 1960s. Thereafter, its use decreased, although it never ceased, indicating that the various models possibly underestimate the risks of persons with longer TSFE.
The findings of several epidemiological studies17 indicate that mesothelioma arises in only a small proportion of persons exposed to asbestos, suggesting that some individuals might have a higher risk than others. Progressive attrition of a high-risk group in the cohort might be responsible for a reduced mortality risk in the longest survivors, who contributed most to the highest categories of TSFE. Nevertheless, this phenomenon, like the other possible sources of bias described earlier, cannot explain the different trends observed for pleural and peritoneal cancer, with a monotonic increase in the latter over time.
Few studies have addressed the long-term patterns of cancer rates after exposure to asbestos. A study of a large cohort of 17,000 asbestos insulation workers by Selikoff and Seidman18 provided findings similar to ours. The mortality rates of pleural cancers (based on 173 cases, among which 55 had a TSFE of at least 40 years) peaked 45–49 years after first employment and declined in the succeeding quinquennium (Fig. 2), whereas the rate of peritoneal cancer (based on 285 cases, among which 111 had a TSFE of at least 40 years) increased steeply over all categories of TSFE18 (Fig. 3). Although based on smaller figures, reduced rates of pleural cancer many years after exposure have also been reported in other cohorts.19–22
Asbestos clearance is a possible biological explanation for the observed late decrease in the rate of pleural cancer. A recent update of the follow-up of the Wittenoom cohort suggested a role of asbestos clearance7, 20; however, on statistical grounds, little significance can be attached to these findings, as only 4 cases of pleural cancer were observed 45 years after first exposure, when the predictions of the different models diverge appreciably. In our study, the elimination model fitted the data significantly better than the traditional model and provided estimates of the coefficients that are consistent with the figures suggested by Berry5, 7 (λ, 7–15% in the various models).
Two assumptions must be made in order to consider cessation of exposure to asbestos as a cause of the observed decrease in rates of pleural cancer: first, cessation of exposure must result in a reversion of the incidence rates toward the background risk of unexposed persons; and, second, asbestos must be effectively cleared from tissues after cessation of external exposure, as occurs for nondurable carcinogens. The former assumption can be derived mathematically if an exact solution of the multistage model is used.12, 23 Asbestos is considered to affect an early stage of a multistage process of mesothelioma carcinogenesis.24 The exact (but not the approximate) solution predicts that, after the end of exposure to an early-stage carcinogen, the incidence rate of exposed persons approaches the incidence rate of those who were never exposed (even if this process is expected to be more gradual for early-stage than for late-stage carcinogens12, 23). Regarding the second assumption, there is evidence from experiments in animals that crocidolite and amosite are eliminated after exposure by inhalation.25–29 There is also indirect evidence, from determinations of the asbestos burden in the lungs of persons formerly exposed to asbestos.30–33 Some authors have postulated that asbestos clearance can explain the observation that the risk for mesothelioma many years after the end of exposure is higher than expected on the basis of the lung content of asbestos and that the latter is consistent with an elimination rate of 9–15% per year,20, 33, 34 an estimate similar to our findings.
In both our study and in the cohort study of Selikoff and Seidman,18 peritoneal cancer rates increased monotonically over time. Should the elimination of asbestos be the explanation for the decrease in rates of pleural cancer, the observed trends in peritoneal cancer could suggest a lesser role, if any, of fiber clearance in the development of this tumor. Few data exist on how asbestos fibers reach the peritoneum and their ultimate fate.35 Therefore it is difficult to speculate on how clearance could affect differently the translocation of fibers to pleural and peritoneal linings. Some studies suggest a different fate for the fibers reaching the pleura and the peritoneum. While short chrysotile fibers are the most common fiber type found in the pleura,36 Dodson et al.35 found long chrysotile fibers in the peritoneum of persons with mesothelioma, suggesting no apparent degradation of the fibers that reach the peritoneal cavity. Moreover, some authors described specific areas in the pleura (but not in the peritoneum), called black spots, where fibers and other particles are trapped and afterward drained through the lymphatic system.37
The observed reduction in mortality from pleural cancer in our study suggests that the number of cases of pleural mesothelioma predicted to occur in the coming decades on the basis of the traditional model14, 15, 38 might be overestimated. The cumulative number of pleural mesothelioma cases observed among miners of Wittenoom in 2000 was 30% lower than that expected on the basis of the traditional model and was predicted to be 50% lower in 2020.7
Our findings, although based on a limited number of cases, suggest that the risk for pleural cancer might not increase indefinitely, whereas it might reach a plateau when a sufficiently long time has elapsed since the start of exposure. This trend might be related to clearance of asbestos from the workers' lungs. Predictions of the number of cases expected in the coming years, assuming a non-zero elimination rate of asbestos, should be considered in addition to those obtained with traditional models. Finally, the differences in the pattern of mortality from pleural and peritoneal cancer suggest that the time trends of these 2 neoplasms should be analyzed separately.
We wish to thank Milena Maria Maule, Franco Merletti, Neil Pearce and Lorenzo Richiardi for useful comments. The study was carried out in the framework of projects partially supported by Special Project Oncologia, San Paolo Foundation/FIRMS and the Italian Association for Cancer Research (AIRC). Francesco Barone-Adesi was partially supported by The Master in Epidemiology, University of Turin.