We found an increased risk of mesothelioma in subjects that were previously irradiated to treat prostate cancer. The risk appeared to increase with latency. Based on the closer proximity to the site of irradiation, one might expect the risk to be higher for peritoneal mesotheliomas. Within the limits of the small number of such cases, this hypothesis appears to be supported—the point estimate for peritoneal mesothelioma risk and EBRT exposure was higher albeit with wide confidence intervals, and the IRRs stratified according to latency presented the same pattern found when analyzing all mesotheliomas combined.
Previous evidence and biological plausibility
Our findings support an association between exposure to EBRT and risk of mesothelioma. This observation is predominantly based on pleural mesotheliomas (accounting for more than 93 % of the cases) which occur distant from the irradiation field for prostate cancer.
Radiation-induced malignancies are usually expected to occur within the irradiated field (e.g., Baxter et al. [
25]). However, even organs far from the irradiated field can still be significantly exposed due to scattered radiation, as well as leakage from the radiation source [
23,
26,
27]. Three-dimensional conformal radiation therapy of the prostate (a frequent treatment during the 1990s [
28]) can expose the pleura to an equivalent absorbed radiation dose up to 25 mSv (Web Table 3); this value is far from being insignificant if we consider that the effective dose for a standard chest radiograph ranges between 0.05 and 0.24 mSv [
29]. When interpreting this value, we should consider that 0.6–1.8 % of the cumulative risk of cancer to age 75 years could be attributable to diagnostic X-rays [
30].
Findings from registry-based studies on second cancers provided inconsistent evidence. On the one hand, previous studies highlighted a possible association between EBRT for prostate cancer and risk of lung cancer [
12‐
14]. On the other hand, a study on second neoplasms after invasive breast cancer did not identify any increase in risk for medium (0.5–1.0 Gy) or low (below 0.5 Gy) doses of radiation [
31]. However, it is interesting to note that among sites receiving high radiation doses, pleural cancers presented the highest point estimate, although based on only two cases [
17].
Our study period was limited (1973–2009), and we studied latencies shorter than those usually reported for asbestos-related mesothelioma [
32]. Nevertheless, previous studies of the latency period of radiation-induced solid tumors suggest an average latency period of 5–15 years, in line with our analysis [
14,
18]. Furthermore, several case reports on EBRT and mesothelioma have described cases occurring after latency periods in the range of 5–41 years [
9].
Epidemiologic evidence that might support an association between exposure to ionizing radiation and mesothelioma is inconsistent. On the one hand, many studies on Thorotrast or EBRT and risk of mesothelioma do report increased risk of mesothelioma among subjects exposed to radiation [
9]. On the other hand, studies among occupational cohorts working in the nuclear industry have not observed increased risk of mesothelioma, at least not that can be confidently attributed to radiation exposure rather than to confounding by asbestos [
9,
33].
Study strength and limitations
Our study was based on a large number of mesothelioma cases, in contrast with previous studies on EBRT and mesothelioma [
9]. Hence, we were able to detect a small increase in risk (about 30 %) and to conduct an analysis stratified according to latency.
The main limitation of our study is the potential for unmeasured confounding as information on personal characteristics and individual exposures was lacking. Confounding due to exposure to asbestos is always a concern when studying mesothelioma. In the present analysis, we were unable to adjust our estimates according to the personal history of exposure. Instead, we used the RR of mesothelioma among males in the county of residence as a proxy measure of exposure to asbestos. Although certainly affected by a high degree of misclassification, this measure of exposure to asbestos was able to capture at least part of the individual risk of mesothelioma, as shown by the well-shaped dose–response relationship. It is important to note that the estimate of interest, that is, the IRR for exposure to EBRT, did not change after the introduction in the multivariate models of the variable for the county’s RR of mesothelioma. This finding suggests that the association between EBRT and mesothelioma was not highly confounded by asbestos in our study population.
It is possible that receiving radiotherapy rather than surgery might in some way be associated with determinants of mesothelioma, although aside from asbestos, there are no established personal risk factors for mesothelioma that could represent a contraindication for surgery. It is possible that there was a higher prevalence of chronic cardiac and pulmonary diseases among people occupationally exposed to asbestos because the asbestos exposure tends to occur more often in lower socioeconomic classes where poorer overall health might increase the aforementioned chronic conditions [
34‐
36]. Treatment decision making in prostate cancer is influenced by the presence of comorbidities; patients affected by chronic diseases have a higher probability of receiving radiotherapy instead of surgery [
37]. Hence, our findings could be at least partially explained by a higher tendency for former asbestos workers to receive EBRT (although we have no direct evidence to suggest this). To explore this possible source of selection bias, we performed a supplemental analysis in which only patients who had received neither surgery nor radiotherapy were used as the comparison group (Web Table 2). This change did not alter the point estimates of the risk associated with EBRT, although the smaller comparison group necessarily resulted in larger standard errors. Reports on the distribution of socioeconomic factors (i.e., level of education and income) highlighted that patients from lower socioeconomic status were less likely to receive any treatments, whereas the percentage of patients receiving radiation was usually independent from socioeconomic status or even showed a positive association [
38‐
42]. Therefore, while the comparison between surgery and radiotherapy could be affected by a bias away from the null hypothesis (i.e., showing a risk greater than the real one), the comparison between patients who received surgery and patients who did not receive therapies should be biased toward the null hypothesis (i.e., showing a risk smaller than the real one). It is also interesting that the IRR for patients who received only surgery compared to untreated patients was 1.00 (Web Table 2); this observation suggests that the distribution of previous occupational exposure to asbestos according to treatment status is not likely to be importantly unbalanced.
We also repeated this analysis after excluding stage I cancers, a subpopulation in which active surveillance has been proposed in the absence of comorbidities [
43]. The risk for patients treated with EBRT compared to patients who did not receive any therapy was still close to that estimated in the main analysis (IRR = 1.28, data not shown). Taken together these findings suggest that selection bias introduced by comorbidities associated with asbestos exposure is not a major concern in our study. Another element supporting the absence of confounding by asbestos exposure is the increase in risk with latency (Fig.
2). Our estimates were adjusted by age and squared age, which were introduced in the models as time-varying covariates; under these conditions, residual confounding by age is unlikely. Moreover, almost no increased risk was observed in the first 5 years after the irradiation. Hence, the latency period from the diagnosis of prostate cancer is unlikely to reflect an age-dependent latency from the first occupational exposure to asbestos. To produce the observed pattern of estimates, confounding by asbestos should act differently in each latency period.
We also performed a target-adjustment sensitivity analysis to explore the difference in prevalence of occupational exposure to asbestos by EBRT status that would have been necessary to explain the observed associations (see Web Appendix 1). In order to completely explain the IRR observed for EBRT, the prevalence of occupational exposure to asbestos would have to have been 30 % higher in subjects exposed to EBRT compared to unexposed subjects (see Web Table 4). And, when applying a more plausible assumption of a latency of 10 or more years, the EBRT group would have to have had 63 % higher asbestos exposure than the comparison group in order to fully explain the observed association between EBRT and mesothelioma. Such large differences, with no evident explanation, are implausible.
Registry-based studies of cancer might be affected by detection bias (also called surveillance bias). Detection bias occurs when there are systematic differences between the study groups in the assessment of the outcome [
19]. This kind of bias is often a serious threat when studying cancers that can be clinically silent for a long period (e.g., prostate cancer or breast cancer). Indeed, clinical follow-up of the primary cancer may elicit the identification of secondary neoplasms that otherwise would have gone undetected. The risk of mesothelioma for subjects treated with EBRT observed in our study increased with latency period; the higher risk was found for latency of 10 years or more. A substantial difference in health monitoring among the studied groups is unlikely to have occurred so far from the primary treatment for prostate cancer. Also, an analysis restricted to patients aged up to 85 years, a subpopulation in which under-ascertainment is less likely in SEER data [
44], produced estimates similar to those obtained when studying the entire population (IRR of mesothelioma 1.24, 95 % CI 1.01, 1.53—data not shown). On balance, we believe that detection bias was not likely to have been a major limitation in our study.
As in some previous analyses of SEER data, we did not adjust by tumor stage and grade. However, a sensitivity analysis confirmed our a priori hypothesis that grade and stage of prostate cancer did not appear to be confounders of the association between EBRT and mesothelioma.
We do not believe that the quality of diagnosis is a serious limitation. Most of the mesothelioma cases were diagnosed with histology or exfoliative cytology which are effective in the differential diagnosis with peripheral lung cancer. On the other hand, no information was provided in SEER on results of an immunohistochemical panel, which has been recommended to distinguish benign from malignant mesothelial proliferations [
45]. We cannot therefore exclude the possibility that the incidence of mesothelioma may have been slightly over-diagnosed; however, this misdiagnosis is unlikely to have been differential with respect to the exposure of interest (beam radiotherapy).
Exposure to EBRT was studied as a dichotomous variable as no detailed information on radiation doses or treatment type was available; therefore, we were not able to study a dose–response relationship. Also, it is likely that there was a certain degree of exposure misclassification because some subjects classified as unexposed at the baseline may have undergone EBRT later in the study period. This source of non-differential misclassification of exposure is likely to have biased estimate toward the null hypothesis.
When modeling on the absolute scale, we found that the increase in risk was modest (3.31 per 100,000 person-years); as a point of comparison, we note that it is smaller than that determined by living in a county at high risk of mesothelioma. This perspective and the very small population attributable fraction (0.49 %) suggest that the contribution of EBRT to the worldwide mesothelioma epidemic has been negligible.