Modelling mesothelioma risk for workers assembling military gas masks

McDonald et al (Occup Environ Med 2006;63:852–5) present an updated analysis of a cohort of mostly women who made gas masks incorporating crocidolite asbestos between 1940 and 1944. They found a high risk of death from pleural and peritoneal mesothelioma, confirming the carcinogenicity of crocidolite, but during the last eight years of follow‐up there were no further deaths from mesothelioma (their fig 1 and table 3).

Figure 1 Predicted mesothelioma deaths, including terms for clearance and death from competing causes.

They noted that the mesothelioma rate (number of mesothelioma per 100 000 person years at risk) did not continually increase with time since exposure, as is often believed, but levelled off after about 30 years. They attributed these observations on the mesothelioma rate to clearance of the fibres from the lung, although they recognised that the absence of deaths in the final years of follow‐up could not be explained by clearance alone.

We agree with the authors that clearance of fibres from the lungs is not a viable explanation for all of their observations. To illustrate this we have modelled the risk with a few simplifying assumptions, in order to look at the time course of the predictions compared to the reported data. We calculated deaths from mesothelioma for a hypothetical group of women exposed to a high concentration of crocidolite for one year starting at age 20, using the approach to estimating risks of mesothelioma outlined by US Health Effects Institute.¹ This model has been widely used to predict future risks for populations exposed to asbestos. We adjusted the multiplicative coefficient (that is, nominal exposure level) to produce the same total number of mesothelioma as reported by McDonald et al (that is, 65 mesothelioma).

Key issues to be taken into account when predicting risks of mesothelioma using standard models are the effects of competing deaths (implicitly adjusted for in observational studies) and clearance of fibres from the lung. Figure 1 shows the effects of these adjustments separately and in combination on the predicted annual deaths from mesothelioma. We adjusted for clearance by including a term to account for clearance of the fibres from the lung as recommended by Berry,² assuming a clearance half‐time of 20 years. Essentially, this model includes an exponential decay function to reduce the predicted annual risk. It represents the hypothesis that there is some progressive elimination of the crocidolite fibres from the lung. Information on risk of death from all causes by age was obtained from the UK Office for National Statistics.

It is clear that without including terms for clearance or competing causes of death the mesothelioma risk increases continuously. With only the clearance term in the model the predicted number of deaths is much lower and tends to plateau at about 1.8 deaths per year. Including competing causes of death rather than clearance causes the predicted number of deaths to peak in about 1990, when the age of those exposed would have been 70 years, and then decline rapidly. The combination of clearance and competing causes of death produces a pattern of predicted deaths from mesothelioma similar to that observed by McDonald and his co‐workers—that is, a rise in the predicted number of deaths, then a relatively steady number of deaths each year between about 1970 and 1990, followed by a rapid decrease in the deaths per year. However, the model still predicts a number of mesothelioma deaths in the latter period of follow‐up.

The inclusion of the elimination function and the competing cause mortality led to a reasonably good agreement with observed data in each period, as shown in figure 2. In this figure, the observed number of mesothelioma are necessarily integers, whereas the predictions are not rounded to whole numbers. The difference between prediction and observed has been rounded to the nearest whole number. Our simplified model confirms that whereas the “elimination factor” does suffice to reproduce the trend over the first eight periods, it does not account for the dearth of mesothelioma in the last eight years.

Figure 2 Number of mesothelioma in the 10 observation periods defined in table 3 of McDonald et al, that is nine five‐year periods from 1956 to 2000, and one three‐year period from 2001 to 2003.

We suggest that there might be at least two possible explanations for the absence of mesothelioma deaths in these later periods. (1) Those surviving into the final eight years might be people who had least exposure among that workforce (for example, they might have been people who left school in 1944 and thus were also relatively young and would therefore be in the final survivors). This would correspond to there essentially being two groups of people in the factory: one with high exposure and a second, younger group, with little or no exposure. (2) Alternatively, the elevated levels of respiratory cancer in the whole workforce might suggest that lung cancer, having a shorter latency,³ claimed the victims before mesothelioma did—that is, some people with high exposure may have been at risk of both diseases and succumbed to the lung cancer earlier than mesothelioma.

We believe that it is important to take account of both clearance and death from causes other than mesothelioma in asbestos risk assessments. Any methods that do not include both these terms will provide an overestimate of risk, particularly for young people exposed to asbestos. Further research is needed to understand whether there are other processes affecting the risk after very long periods of follow‐up.