Incidence of Leukemia, Lymphoma, and Multiple Myeloma in Czech Uranium Miners: A Case–Cohort Study

Uranium mines in the Příbram region of the Czech Republic, about 60 km southwest of Prague, were the largest in the former Czechoslovakia. The employment registry of the Příbram Uranium Industry (UI) lists nearly 48,000 employees who ever worked in the Příbram mines during the operating period (1949–1991). Our cohort consisted of male workers who 1) were listed in the UI employment registry at any time between 1 January 1949 and 31 December 1975; 2) had registry records indicating that they had worked underground for at least 1 month before 1 January 1976; and 3) lived in the former Czechoslovakia on 1 January 1977. All 27,441 workers who satisfied criteria 1 and 2 were matched against the National Registry of Inhabitants and the National Pension Funds registry to check their vital status as of 1 January 1977. The matching success rate was 92%. The study cohort included 23,043 male workers who satisfied criteria 1–3. The study protocol was reviewed by the chair of the Institutional Review Board (IRB) at the NIEHS and determined to be exempt from full IRB review because it involved linkage of existing records and analysis of coded data without identifiers.

Complete exposure and follow-up histories were determined for 177 cases and 2,393 subcohort subjects. The total study population was 2,558 because of a small overlap between the cases and the subcohort. Table 1 shows the cohort, the subcohort, and the cases classified according to the sampling strata. Most workers in the cohort were ≤ 45 years of age in 1977. Because of stratified sampling with respect to age, the subcohort included relatively older subjects. The mean age ± SD of the study subjects at the start of follow-up was 50 ± 10.6 years. The median follow-up duration was 18.8 years.

Among hematopoietic malignancies, the most frequent diagnosis (Table 2) was CLL, followed by non-Hodgkin lymphoma, multiple myeloma, Hodgkin lymphoma, and other diagnoses. The average age at diagnosis was 60.1 years (range, 28.4–86.4 years).

The mean ± SD duration of employment was 6 ± 6.6 years (range, 1 month–39 years). Over 46% of the workers began employment before 1955; < 10% were still employed at the start of follow-up. The mean ± SD lifetime radon exposure was 64.1 ± 98 WLM (median, 23.2 WLM; range, 0–959 WLM). Annual radon exposure rates decreased over time because of improved ventilation and other protective measures. The mean annual exposure rate was 29.3 WLM/year in 1949–1955, 11.7 WLM/year in 1956–1965, 2.9 WLM/year in 1966–1975, and 0.7 WLM/year in 1976–1992. Consequently, younger subjects had much lower radon exposures than older subjects (Table 1). Over 69% of the subjects were classified as moderate and heavy smokers.

Table 3 shows observed numbers of cases of each type of cancer by radon exposure category, along with the ratio of observed numbers and expected numbers under the assumption of no association of incidence with exposure. Table 3 also presents incidence rates standardized to the age distribution of the whole cohort. For all hematopoietic cancers combined, there was no consistent pattern in incidence rates with increasing exposure. For all leukemia types combined and for CLL, the rates were not entirely consistent but generally rose with increasing WLM.

In proportional hazards analysis, the risk of all leukemias combined and the risk of CLL specifically was significantly associated with radon exposure (Table 4). To evaluate the associations, we transformed the power risk parameters to reflect the risks for the upper exposure quintile (110 WLM) relative to the lower quintile (3 WLM). These estimated risks were 1.75 for leukemia and 1.98 for CLL, with CIs excluding 1. The RR for Hodgkin lymphoma and myeloid leukemia was also elevated but was not statistically significant. Non-Hodgkin lymphoma and multiple myeloma were not associated with radon. The analysis of all lymphocytic leukemias combined is not presented because this group included only 3 non-CLL cases. Separate analyses of ALL and acute and chronic myeloid leukemia were not performed because the number of cases was too small.

We used the grouped Poisson model to estimate linear ERR parameters to compare the power risk model to the linear ERR model. RRs from the linear ERR model were similar but slightly attenuated. For example, the estimated RR comparing 110 WLM to 3 WLM was 1.46 for all leukemia combined and 1.63 for CLL (results not shown). Figure 1 shows that the power model gives higher radon risk estimates for CLL than the linear ERR model but has smaller residuals at the higher exposure groups. Maximized Poisson pseudolikelihoods for both models were very close to each other, indicating that the goodness of fit was about the same.

Possible confounding and effect modification by other factors was also investigated. Calendar year of follow-up, age at the start of exposure, exposure rates, and year of start of employment did not affect the incidence rates or modify the effects of radon exposure. However, we found that the association of radon with leukemia and CLL significantly decreased with time since exposure. For example, the estimated RR for CLL comparing 110 WLM with 3 WLM was 1.15 (95% CI, 0.68–1.94; p = 0.61) for exposures acquired > 25 years ago, 1.93 (95% CI, 1.17–3.19; p = 0.024) for exposures acquired 15–25 years ago, and 4.20 (95% CI, 1.84–9.61; p = 0.02) for most recent exposures (2–15 years ago).

We found a significant relationship between cumulative radon exposure and incidence of leukemia, especially CLL. For myeloid leukemia and Hodgkin disease, the risk was elevated but nonsignificant. We saw no association of radon with the incidence of non-Hodgkin lymphoma or multiple myeloma.

Miners are also exposed to gamma radiation, which is associated with non-CLL leukemia. In the mines, gamma radiation and radon exposure are correlated, and the observed risk associated with radon could be explained by exposure to gamma irradiation. In our study, the estimated lifetime gamma exposures were relatively low (mean ± SD, 10.3 ± 14.3 mGy; range, 0–227 mGy) and highly correlated with cumulative radon exposures (correlation 0.80 on the log scale). We did a parallel analysis of gamma exposure and found a significant association with all leukemias combined (RR = 1.63; 95% CI, 1.05–2.54; p = 0.03) and CLL (RR = 1.96; 95% CI, 1.12–3.42; p = 0.02) when comparing exposures in the 80th percentile (20 mGy) to the 20th percentile (1 mGy). This result seems implausible at this low dose, although we cannot rule out such an effect, perhaps in combination with an effect due to radon. There are two potential alternative explanations: either the elevated risk with gamma reflects the strong correlation with radon, or our estimates of pre-1966 gamma exposures were too low. We have no data to verify our ore-productivity–based estimates of gamma exposures before 1966. It is unlikely that non-radiation exposures were responsible for the observed risk, because the workers in Příbram were not exposed to any other substances suspected of causing leukemia.

Smoking was not associated with any hematopoietic malignancy and did not appear to confound the association between radon and leukemia. The estimated RRs due to smoking ranged from 0.83 to 1.72 with wide CIs. Smoking effects could not be evaluated with much precision because smoking was very common and our data on smoking were limited.

The primary strengths of this study are a) the use of incidence rather than mortality as the outcome; b) the completeness and quality of the national cancer registries; c) the availability of detailed work histories; d) radon exposure data based on almost 250,000 radon measurements in the workplaces; and e) statistical methods that do not lose information by grouping age and exposures. The case–cohort design allowed an exceptionally detailed evaluation of exposures based on employment logs of the cases and subcohort members. Beginning follow-up after the end of the exposure period for most miners made this study particularly suitable for detecting long-term effects of exposure.

This study has several limitations, although none seem to account for the observed association between radon exposure and leukemia. We had no information on cancer diagnoses before 1977 and may have missed some cases of ALL, in particular. Furthermore, someone with leukemia diagnosed before 1977 could have entered the study as a subcohort control. The incidence of CLL would be affected more than that of other cancers because of the long survival of CLL patients. Such subjects would be older, on average, than our study cohort, and their radon exposures would be higher. Thus, including many unrecognized pre-1977 CLL cases would bias the observed effect of radon on CLL downward. In addition, diagnosis of CLL can be delayed because the disease often remains asymptomatic for a long time. However, all UI workers underwent annual medical examinations (including blood analyses) during their employment and were also examined, although less frequently, in retirement. Thus, it is unlikely that a substantial number of unrecognized CLL cases were included in the subcohort.

The exposure measurements are subject to measurement error, which is largest for high exposures acquired in the early period. The analyses are not adjusted for exposure measurement errors. These errors typically attenuate an existing exposure effect rather than induce an effect where there is none. It is reassuring, however, that the exposure effects were strongest in the most recent period. The conclusions hold even if the high exposures acquired in the early mining period are eliminated.

Many studies of radiation–cancer associations fit the ERR model with a linear dose–response relationship. We chose the power RR model because the linear ERR model tends to suffer from unreliable standard errors and poor power (Breslow and Day 1987). The power RR model does not share the problems of the ERR model, can be used with ungrouped data, and provides better power and more reliable p-values and CIs. The goodness of fit was similar for both models.

In the past, radon was not considered a potential cause of leukemia because it was believed that most of its alpha activity was released in the respiratory tract. However, several authors (e.g., Henshaw et al. 1990; Richardson et al. 1991) have suggested that radon inhalation may deliver a non-negligible alpha radiation dose to bone marrow and thus increase leukemia risk. Inhaled radionuclides may transmit their alpha activity directly after reaching bone marrow via the blood stream. Because of its high solubility in fat, radon gas itself may concentrate in bone marrow fat cells and emit alpha particles to the surrounding hematopoietic stem cells (Allen et al. 1995). Recently published models that simulated radiation dose to red bone marrow after inhalation of radon attributed a much more important role to radon gas than to radon progeny (Kendall and Smith 2002). Thus, it is plausible that radon causes much of the observed increased risk of leukemia.

Our findings challenge the idea that CLL is not radiation related. Although well accepted, the evidence supporting this notion is not persuasive (Richardson et al. 2005). CLL etiology is still unknown, but the pathogenesis of CLL appears to be similar to other leukemia subtypes. Genetic alterations, inherited or acquired, are likely to lead to any type of leukemia. Like other subtypes, CLL is associated with specific chromosome and immune system alterations. Altered apoptosis is likely to be involved in CLL leukemogenesis. Mutations of tumor suppressor genes, especially p53, and an upregulated expression of the bcl-2 gene may have an important role in the inhibition of apoptosis (Irons and Stillman 1996; Stilgenbauer et al. 2002). Ionizing radiation produces biologic damage that may play a role in one or more stages of the neoplastic process leading to CLL. There is no evidence from research at the molecular level that CLL cannot be induced by ionizing radiation.

CLL is the most frequent type of leukemia among adults in Western Europe and North America. It is extremely rare in East Asia. Therefore, studies of the Japanese atomic bombing cohorts had limited power to detect an association with CLL. CLL may remain asymptomatic for years. Patients often have a prolonged, nonaggressive course marked only by a stabilized lymphocytosis in the peripheral blood and bone marrow. It is not unusual for CLL to be revealed by chance during a consultation for apparently unrelated disease or during a routine examination. Many patients with CLL require no specific treatment and die from other diseases typical for old age (Greer et al. 2003; Henderson et al. 2002).

Given the characteristics of CLL, a substantial number of cases will be missed in studies that rely on death certificates. The accuracy of death certificates compared with hospital and autopsy diagnoses in the same cohort has been shown repeatedly to be poor, especially for diagnoses with low case-fatality. For example, Demers et al. (1992) compared death certificates and autopsy data and found that only 2 of 24 autopsy-diagnosed cases were reported on the death certificates. In another study, Percy et al. (1990) compared underlying causes of death based on death certificates with hospital diagnoses; only 65% of death certificate-based diagnoses were consistent with hospital records. In a study of deaths among atomic bomb survivors, the overall agreement between death certificates and autopsy diagnoses was only 52% (Ron et al. 1994). Thus, mortality studies investigating diseases with low fatality rates, such as CLL, are likely to miss exposure effects because of poor ascertainment of cases and competing causes of death. All previous studies of uranium miner cohorts collected only mortality data based on death certificates.

To overcome the limitations of mortality studies, we studied incidence of hematopoietic malignancies in uranium miners and found a significant association between cumulative radon exposure and incidence of all leukemias, especially CLL. Our findings support the hypothesis that CLL is also radiogenic (Richardson et al. 2005). If this conclusion is confirmed by future studies, leukemia, including CLL, should be considered an occupational disease in workers with prolonged low-level exposure to radon and perhaps to other kinds of radiation.

We thank the Health Institute of the Uranium Industry and the Uranium Industry Concern for providing exposure data and facilitating access to records maintained by government agencies.

This work was supported through the Intramural Research Program of the NIEHS and by a contract from the NIEHS awarded to the Center of Epidemiological Studies.