Familial aggregation of malignant mesothelioma in former workers and residents of Wittenoom, Western Australia



Clustering of cases of malignant mesothelioma within families has often been observed, but disentangling genetic and exposure effects has not been done. Former workers and residents exposed to crocidolite at Wittenoom, Western Australia, where many families shared exposure to asbestos, have had high rates of mesothelioma. Our study aimed to estimate the additional risk of mesothelioma in relatives, after allowance for common exposure to crocidolite. More than 11,000 former asbestos workers and residents from Wittenoom have been followed up in cancer and death registries. Levels of exposure for all members of the Wittenoom cohorts have been estimated previously. Relationships between family members of all mesothelioma cases were established from questionnaires, birth and death certificates. Expected numbers of cases of mesothelioma were estimated by fitting a Weibull survival model to all data, based on time from first asbestos exposure, duration and intensity of exposure and age. For each family group, the earliest case was considered the index case. Predicted risk was estimated for each subject from the time of diagnosis of the index case. Familial risk ratios were estimated by dividing observed cases by the sum of risks of all same degree relatives of index cases. There were 369 family groups with at least one case of mesothelioma and a further 25 cases of mesothelioma among relatives in the same families, with 12.9 expected. The risk ratio for blood relatives was 1.9 (95% confidence interval [CI] = 1.3–2.9, p = 0.002). These findings suggest an important, but not large, genetic component in mesothelioma, similar to many other cancers.

Genetic factors in cancer have long been of interest and there is direct evidence of a genetic component in many forms of cancer. Most cancers have a great amount of worldwide variation in incidence that could be owing to racial differences as well as environmental differences and there is also strong evidence that genetic susceptibility plays a large part in many common cancers.1

The likelihood of inherited predisposition to malignant mesothelioma (MM) was first raised in 1978.2 When familial clustering of cancers occurs, it is difficult to disentangle the effects of the shared environment and the shared genes of the families. This is made more difficult in the case of asbestos and MM because of the likely increase in exposure of families of workers at home,3, 4 where carrying asbestos fibres into the home on work clothes has often been reported, and in the neighbourhood of asbestos works; most reports of clustering of MM are linked to definite exposure to asbestos at work or in the home.5 Other studies have implicated a high genetic susceptibility to MM in Cappadocia where families have been exposed to erionite,6 whereas a study of mesothelioma registries in Italy found little evidence of familial clustering among 1,954 cases of MM.3

Increased rates of death from asbestos-related diseases have already been reported in former workers and residents exposed to crocidolite at Wittenoom, Western Australia.7, 8 Disentangling the effects of common exposure to asbestos from common genetic associations is rarely possible, but the Wittenoom subjects and their families, because of their detailed follow-up over many years and their exposure assessments, provide a unique opportunity to study the interaction of genetic and environmental changes which result in asbestos-related diseases (particularly MM, asbestosis and lung cancer).

The aim of our study was to estimate the additional risk of MM in relatives, after allowance for common exposure to crocidolite.


Crocidolite (blue asbestos) was mined by Australian Blue Asbestos at Wittenoom Gorge in Western Australia from 1943 until 1966. Miners and millers who worked for the local asbestos company and their relatives and others who lived in the local township have been the subject of many epidemiological studies. Assembly of both cohorts has been described elsewhere.7, 9 The overall cohort with sufficient data for these analyses consisted of 6,694 former miners and millers from Wittenoom together with 4,518 residents of the town of Wittenoom, who did not work at the mine or mill. Although both cohorts were used to obtain the statistical model for estimating risk of MM from exposure to asbestos, only families with at least one case of MM were used to estimate the additional risk from having a diagnosed family member.



Mortality was assessed by linking the cohort to the National Death Index through the Australian Institute for Health and Welfare (AIHW) and the Western Australian Registrar General for births, deaths and marriages. Cancer incidence was obtained from the National Cancer Statistics Clearing House through the AIHW, the Australian Mesothelioma Registry and the Western Australian Cancer Registry, including the Western Australian Mesothelioma Registry. The largest number of migrants from non-English–speaking countries working at Wittenoom was from Italy. Untraced subjects with an Italian name have been searched in Italy and their vital status and incidence of MM have been assessed.10 Although mortality follow-up may have missed a proportion of subjects who have died, we are unlikely to have missed any MM cases because of intensive State and national surveillance programs.

Ethical approval for the study was provided by the Human Research Ethics Committee of the University of Western Australia.

Exposure assessment

Characteristics of the two cohorts and assessment of exposure have been reported previously, workers generally only remaining a short time but residents staying longer.7, 11 Dust concentrations, particularly in the old mill (prior to 1958), were high, and estimated cumulative exposures (in fibres/millilitre years) calculated from them (the sum of the product of duration and level of exposure for each period of exposure) were shown to be consistent with measures of lung fibre burden.12 Each subject thus had two measures of exposure: duration (the log of total days of exposure) and either cumulative exposure (fibres/millilitre years) or average or intensity of exposure (the log of [cumulative exposure/years of exposure]). Workers had both environmental and occupational exposures combined. Both measures were counted at the completion of exposure, because average exposure hardly changed throughout each person's time at Wittenoom, and most of the cohort (>97%) were exposed for <10 years and, as shown by de Klerk et al.,13 for most realistic values of exposure, the effect of modelling duration of exposure as complete merely has the effect of reducing the time exponent p (see further below).

Family relationships

Relationships between family members were established from birth, death and marriage certificates, and from detailed questionnaires provided to all cohort members who have been traced. Family groups were formed consisting of all cases of MM and their relatives, and were stored in a separate data file. The earliest case of MM was counted as the index case in each family.

Data analysis

A Weibull survival model based on time from first exposure to asbestos, duration and intensity of exposure, and age at earliest exposure, was fitted to all data (workers and residents), because Weibull regression estimates the effects of the underlying time variables and thus easily enables predictions of individual risks.14 We used the accelerated failure time version of the Weibull model, so that negative coefficients, because they imply a shorter time to failure, translate to increased hazard ratios and vice versa. Fractional polynomials were used to examine the linearity of the associations between any of the continuous variables (duration and intensity of exposure and age at first exposure) and incidence of MM.

Follow-up time for each person was from date of first exposure to the earliest of: end of follow-up (December 31, 2006), death or diagnosis of MM. This model enabled calculation of predicted risk of disease at any time from first exposure for all 11,212 subjects.

Using the results from this analysis, we were then able to estimate individual risk within the second data set, consisting of all families with at least one case of MM, in the following way.

The probability S(t) of surviving to time (or age) t without MM was estimated by:

equation image(1)

where p is the Weibull “shape” parameter, a is the constant in the regression equation, and b1, b2,… are the coefficients for the covariates x1, x2,… (e.g., log[exposure intensity] and log[exposure duration]). The risk of MM for each person at any given t is then given by R(t) = 1 − S(t) and the cumulative risk (CR) of MM between two times, t1 and t2, is given by:

equation image(2)

The expected number of cases was estimated by calculating the risk for everyone in each family group, apart from the index case, from the time of diagnosis of the index case to the end of follow-up of each relative, using R(t2) − R(t1), (or S(t1) − S(t2)), as above, where R(t2) is the risk from first exposure until the end of follow-up, and R(t1) is the risk from first exposure until diagnosis of the first or index case in the family. A familial risk ratio (analogous to the sibling risk ratio [λs]) was then estimated by dividing observed familial cases by those expected based on the sum of the risks of all relatives of the index cases as calculated above. The summations were done separately for first degree (parent, sib and child), second or higher degree (aunt, uncle, cousin, grandparent, etc.), and spouses.

The significance of the risk ratio was assessed using the one-sided probability of obtaining the observed number of cases (or more) from a Poisson distribution with mean equal to the expected number and 95% confidence intervals were also based on exact values using the same Poisson assumption.15

In order to examine the likely variability in risk ratios arising from variability in the Weibull model coefficients and from any assumptions about the distribution of (O/E), we used a bootstrap procedure16 with the family data set. Bootstrapping is very useful for estimating statistics arising from complex procedures such as done here and providing a realistic estimate of their variability.17 It can be a quite powerful tool in this kind of situation. First, a bootstrap sample was selected (a bootstrap sample is a sample selected at random from the data set under analysis but with replacement, so that the sample size is the same as the total sample, but because of the random sampling with replacement, many of the subjects may appear in each sample several times and some not at all). Then for each subject in the sample, a random selection was made from each Normal distribution with mean and standard deviation equal to the coefficients and standard errors of each of the terms in the risk prediction model (Results section and Table 3), and then the subject's risk was estimated using the values of these random selections in Eq. (1). For each bootstrap sample, the observed cases were summed, the predicted risks were summed and O/E was calculated. This process was repeated 1,000 times and the mean and 2.5 and 97.5 percentiles were used to provide, perhaps, more realistic estimates of relative risks and confidence intervals.

It is also worth noting that using the Weibull regression model as described above, the hazard function (and thence the incidence rate) is proportional to time since first exposure (t) to the power (p − 1), where (p − 1) has been found to be between 3 and 4 for MM.5, 18, 19 The model also included no specific lag period, a time period during which no cases can possibly occur, as the probability of any cases occurring within 10 years of exposure using the above model is very low anyway and a continuous model is more readily interpreted in terms of most cancer induction models.20


There were 11,212 subjects with complete data available for estimating the MM incidence model (Table 1). Of these, 1,022 were in families with one case of MM and 142 in multiple case families. As expected, the complete cohort had a higher percentage of males and workers and a shorter duration of exposure than the family cohorts. The average intensity of exposure was lower in the family cohorts, because of the lower intensity of exposure in the ex-residents and females.

Table 1. Study subjects: demographic and exposure characteristics (medians and [10th and 90th percentiles])
inline image

For family groups there were 369 index cases of MM and 178 of these had no known relatives (and hence did not contribute any family risk or expected number information to the analysis below), and hence there were 191 families with more than one member, 20 families with more than one case (Table 2). As larger families contribute greater potential for familial risk through greater person-time at risk, the multiple case families were larger. As would be expected, the familial cases had higher average values of all predictor variables, as well as a longer time from first exposure to diagnosis (Table 1).

Table 2. Family sizes in mesothelioma family cohort
inline image

There were a total of 396 cases of MM up to the end of follow-up time (December 2006) for the whole cohort. Incidence of MM was strongly dependent on time from first exposure (around the third power (i.e., 3.94 − 1) as found previously (e.g. Ref. 21), as well as the log of average exposure to asbestos and the log of days of exposure (Table 3). Two polynomial terms for age at first exposure (1/age2 and log[age]) were also found to fit the data well (and much better than a linear age term, p < 0.001), indicating a steady less than linear increase in risk after about 6 months of age. After including these two terms and the sex effect, we found no additional significant difference between workers and residents as had been suggested earlier for lung cancer.22

Table 3. Coefficients1 in Weibull accelerated failure-time model (combined cohort)
inline image

These coefficients, together with each family member's values of those variables, were used to calculate individual risks (using Eqs. (1) and (2) above, as partly shown in Fig. 1), and summed to obtain the expected or predicted number of cases, a total of 12.9, with a total of 25 observed familial (related) cases. None of these cases had any other cancers. Observed-to-expected ratios were significantly different from unity for both first- and second-degree relatives although the ratio was slightly higher in second- than first-degree relatives, but not significantly so (Table 4). The relative risks were slightly lower using the bootstrap procedure, but confidence intervals remained similar. Between spouses, the ratio was <1, with wide confidence intervals under either estimation procedure.

Figure 1.

Example of calculation of risk of mesothelioma after starting work age 20 for 1 year at exposure of 10 f/mL. Index family case diagnosed when subject aged 50, subject followed up to age 62, and hence expected risk = 0.0099 − 0.0026 = 0.0073.

Table 4. Observed and predicted familial mesothelioma cases
inline image


The question of heritable factors in the development of MM (and other cancers) has been described as being of fundamental relevance4 to the understanding of the pathogenesis of cancer. Because of the rarity of MM in people who have not been exposed to asbestos (or asbestiform fibres) MM is an ideal model for aetiological investigation. This is the first study that has been able to adjust properly for degree and timing of asbestos exposure when evaluating genetic risk of MM, indicating an approximate doubling in risk for relatives compared to no increase in spouses although based on small numbers.

The Wittenoom cohorts have had well-characterized exposures, almost exclusively to crocidolite, in amounts that have been estimated from airborne dust measurements conducted in the industry during its operation from 1943 to 1966 and in the township until its closure in 2006/2007. Follow-up is considered virtually complete, at least for cases of MM, as continuous searches have been conducted in Australia and Italy since the establishment of the cohorts in 1975 (workers23) and 1989 (residents9) and cancer has been a notifiable disease in Australia since 1981. All cases included in our study have been confirmed pathologically (including by cytology) and must have also been confirmed by mesothelioma registry committees across all states of Australia to ensure accuracy of diagnosis.

This analysis is a first and approximate look at estimation of a genetic component in MM among these families. Although, ideally, assembling pedigrees of all families, even those with no cases of MM, might improve estimation of the underlying asbestos-associated rates, the amount of work involved did not seem justified at this early stage. Further refinement of the method could involve an iterative process of re-estimating the risk model on the combined cohort but incorporating the estimated familial effect, then re-estimating the familial effect and so on. Alternatively, more complex models incorporating within family effects could be fitted and genetic effect estimated, but again, knowledge of all pedigrees would be required.24 However, inclusion of family members who had never been at Wittenoom and had no asbestos exposure was not required as they would have added essentially zero risk to the expected numbers and zero to the observed numbers.

Although the bootstrapping procedure provides some indication of likely variability in our relative risk estimates caused by variability in our data, there are many other sources of possible bias that could lead to error in our estimates. All family members were assumed alive to the end of follow-up unless known to be dead or diagnosed with MM. This could have biased the expected numbers upwards and hence the relative risk (RR) downwards, but the effect is likely to be small, as most losses to follow-up in these cohorts have occurred immediately after leaving Wittenoom.7 There may be subjects who migrated out of Australia (whether Australian born or migrants from overseas) to countries other than Italy, so that we could have missed cases as well as over-estimating person-time through missed deaths from other causes. Inaccuracies in exposure estimation, particularly at lower levels of exposure, could have led to mis-specification of the exposure–response relationship, with an unknown effect on estimating the expected numbers. The lack of a significant difference between workers (occupationally exposed) and residents (environmentally exposed) in the prediction model suggests that such an effect may not be too large.

Our findings suggest that there is an important, but not large, genetic component in MM similar to that found for other cancers.25 This result indicates a doubling of risk that may be attributable to genetic factors and supports the contention of other studies that were based on less precise exposure information in case–control studies26, 27 or descriptive data.4 The difference in relative risks between spouses and relatives also indicates that familial clustering in MM is unlikely to be caused solely by common infective agents such as SV40 virus28, 29 or by common dietary factors.30, 31

As all subjects were exposed only to crocidolite, extension to other forms of asbestos might not be justified. Although there has never been any evidence that MM caused by one type of asbestos is different in any way from MM caused by another type,32 individual responses to different asbestos or fibre types could be under different genetic control.

The results of our study support the efforts to explore somatic molecular genetic alterations in patients with MM as commenced by us33 and others.34 MM is a good human model to study because of its almost exclusive association with asbestos or erionite exposure, apart from rare findings of association with ionizing radiation, for example Ref. 35.