Cancer Risk From Occupational Exposure to Ionising Radiation
Cancer Risk From Occupational Exposure to Ionising Radiation
This study provides evidence of a linear increase in the excess relative rate of cancer mortality with increasing exposure to ionising radiation at the low dose rates typically encountered in the nuclear industries in France, the UK, and the USA. Restricting analyses to information regarding doses below 200, 150, and 100 mGy showed that the estimated excess relative rate per Gy for all cancers other than leukaemia were not driven by the highest dose categories. Analyses restricted to these lower doses also address the radiation protection community's interest in epidemiological evidence of a radiation dose-cancer association in these low dose ranges. INWORKS thus provides supportive evidence for a positive association between radiation dose and all cancer other than leukaemia, even if less precise when analyses are restricted to data for the 0–100 mGy dose range.
The primary basis for the radiation risk estimates used to establish contemporary radiation protection standards comes from analyses of cancer in the Life Span Study of the Japanese atomic bomb survivors. Historically, it has been assumed that radiation-solid cancer associations diminish with falling dose rate. For example, the International Commission on Radiological Protection (2007) recommended that regulators divide the radiation risk coefficients obtained from the study of Japanese atomic bomb survivors in half when estimating risks for cancers other than leukaemia in settings with exposures of low dose and low dose rate radiation. This was recommended because the high dose rate exposures from the atomic bombings were assumed to be more dangerous than the low dose rate exposures typically encountered by workers and members of the public. Questions about the effects of such low dose exposure have, in part, motivated studies of nuclear workers since the 1970s.
Our estimated association between radiation and solid cancer (excess relative rate 0.47 per Gy; 90% confidence interval 0.18 to 0.79) is larger than but statistically compatible with the estimate from a mortality analysis of male survivors of the Japanese atomic bomb exposed at ages 20–60 years (excess relative rate 0.32 per Sv; 95% confidence interval 0.01 to 0.50). Statistical compatibility of risk estimates between studies may suggest some degree of coherence in the evidence derived from these large studies. However, in observational cohort studies, such as INWORKS and the Life Span Study of Japanese atomic bomb survivors, large sample sizes and statistical precision are no protection against bias. We have attempted to deal with some concerns regarding bias through decisions in study design. To this end, INWORKS was not intended to assemble the largest number of nuclear workers possible, but rather to assemble those cohorts that were most informative with regard to quality and completeness of exposure and follow-up data.
The parent study of the INWORKS collaboration included 407 391 workers from 15 countries. Although INWORKS included fewer nuclear workers than the earlier study, dose-response analyses in INWORKS encompassed substantially more cancer deaths than the parent study (17 957 v 4770 solid cancers), reflecting the extended follow-up of the INWORKS cohorts. INWORKS did not include data from Canada, a cohort for which the excess relative rate per Gy estimate was considerably larger than that observed in most other countries in the parent study, and for which concerns have been raised regarding data quality and completeness. In our analysis, no single country's data exerted a large impact on the magnitude of the summary risk estimate. Rather, a statistical test of heterogeneity by country rejected the conclusion of significant variation in the radiation-cancer association between the three countries.
The summary risk estimates in the current analysis are more precise and are larger in magnitude than those obtained from previous country specific analyses. For example, a previous analysis of data from France reported an estimated excess relative rate for solid cancers of 0.34 per Gy (90% confidence interval −0.56 to 1.38). Data from the UK National Registry for Radiation Workers had an estimated excess relative rate for death due to cancers other than leukaemia of 0.28 per Gy (0.02 to 0.56). Finally, data from the US showed an estimated excess relative rate for cancers other than leukaemia of 0.14 per Gy (−0.17 to 0.48).
Our estimated radiation risk coefficients could be somewhat larger than those in previous analyses of the constituent cohorts because we adjusted the recorded dose to account for bias in historical dosimeter response and attenuation, taking the estimated colon dose as the quantity of interest. Analysis of the INWORKS data using recorded photon dose as the dose metric, rather than adjusted estimates of colon dose, yielded somewhat lower estimates of association, although use of adjusted colon dose estimates resulted in no significant improvement in model goodness of fit (web Table A3). This finding accords with the general principle that it is possible to use assumptions about patterns of exposure misclassification (for example, different exposure periods or conditions) to reduce bias, but it is much more difficult to recover the precision that would be obtained if one knew each person's true exposure.
Our adjusted dose estimates drew on the substantial work done to characterise the performance of the various radiation dosimeters used in France, the UK, and the USA over the study period and account for differences between countries and over time in dosimeter performance. Use of colon dose estimates facilitated comparison of our radiation risk estimates with those reported in mortality analyses of Japanese atomic bomb survivors that also related to estimated colon dose. However, exposure measurement errors related to personal dosimeters, monitoring practices, and historical records, particularly in the early years of operation, remain a study limitation. Radiation exposures might also have occurred outside of employment at facilities for which we have dosimetry records, and some workers could have had occupational radiation exposures that were not identified in the records available for this study.
In view of our focus on mortality due to cancer, a reasonable concern is potential confounding by cigarette smoking, which was unmeasured in our study. Contrary to the pattern that would be expected if there was confounding by smoking, the magnitude of the estimated excess relative rate per Gy under a 10 year lag was essentially unchanged after excluding lung cancer. A separate paper on non-cancer disease further supports the conclusion of no significant confounding by smoking as evidenced by the lack of association between radiation dose and chronic obstructive pulmonary disease, an outcome strongly associated with smoking. Although there has been interest in the joint effects of radiation and smoking on cancer risk, this could not be evaluated in our study.
Similarly, we could not directly adjust for the effects of exposure to other known occupational lung carcinogens, such as asbestos. However, the magnitude of the estimated association between radiation dose and solid cancer mortality remained unchanged after the exclusion of lung cancer (and further exclusion of pleural cancer). Therefore, occupational lung carcinogens may not be an important confounder in the overall analysis of the association between radiation and solid cancer.
In INWORKS, adjustment for socioeconomic status reduced the magnitude of dose-response estimates, suggesting positive confounding by socioeconomic status. This variable, which is primarily based on job title, is likely to be a poor proxy for factors that relate social class to mortality differences, and suggests the possibility of residual confounding of radiation-cancer associations by socioeconomic status. We adjusted for duration of work owing to evidence of a modest deficit in relative mortality among workers who had at least 10 years of radiation work, and a larger deficit in relative mortality among workers who had at least 30 years of radiation work. Stratification by duration of radiation work slightly increased estimates of association, suggesting negative confounding due to preferential retention of workers in better health (sometimes termed healthy worker survivor bias). We assessed the sensitivity to adjustment for these variables by fitting a simpler model that adjusted only for country, age, sex, and birth cohort. The estimated association between dose and cancer mortality other than leukaemia was similar in magnitude and precision to that obtained from the fully adjusted model, suggesting that the net effect of adjustment for these variables was small (web Table A4).
In the international collaborative study of cancer risk among radiation workers in the nuclear industry, people with potential exposure to neutrons, which was difficult to reliably quantify using the historical personnel dosimeters, were excluded from analyses to focus on workers with well-measured photon doses. A concern raised regarding this exclusion was that it excluded a large number of workers with high cumulative external photon doses.
In the current analysis, we included workers with potential exposure to neutrons and adjusted where possible for neutron monitoring status. In sensitivity analyses, we excluded the 13% of the cohort that ever had a recorded neutron dose. The resulting estimated association between colon dose and mortality due to cancer other than leukaemia (excess relative rate 0.55 per Gy; 90% confidence interval 0.17 to 0.95) was similar to that obtained for the whole cohort after adjustment for neutron monitoring status (0.48 per Gy; 0.20 to 0.79). However, owing to the limitations of historical neutron dosimetry information, an analysis restricted to workers with no recorded neutron dose is likely to have included workers who had unrecorded neutron exposures, particularly among those employed in the early years of operations.
The estimated association among workers with a positive recorded neutron dose (excess relative rate 0.36 per Gy; 90% confidence interval −0.08 to 0.88) and those with a recorded neutron dose exceeding 10% of total dose (0.62 per Gy; −0.50 to 2.09) were also similar in magnitude to, but less precise than, the whole cohort estimate after adjustment for neutron monitoring status. Therefore, the summary adjusted estimate does not appear to have obscured any meaningful heterogeneity in the excess relative rate per Gy by neutron monitoring status (web Table A5).
Employees who had recorded neutron doses—which reflects work in radiologically controlled areas where neutron dosimeters were issued—tended to have lower baseline rates of cancer mortality than those who did not. Reduced mortality rates among workers in radiologically controlled areas has been attributed to factors such as restrictions on smoking in such areas, and additional medical screening for work in areas where additional hazards might occur. Adjustment for neutron monitoring status accounts for such differences in baseline rates between groups, yielding a summary adjusted estimate comparable to the stratum specific estimates of excess relative rate per Gy. However, an unadjusted estimate produced a smaller value (0.20 per Gy; 90% confidence interval −0.03 to 0.46; web Table A5). Adjustment for neutron monitoring status, however, might be inadequate to fully control for differences in baseline rates between these groups, owing to limitations of historical neutron dosimetry information. Furthermore, bias could persist in adjusted analyses if health related selection out of employment affects a worker's future monitoring status and exposure history, and is itself affected by previous radiation exposure.
INWORKS included workers with potential for committed doses from incorporated radionuclides. As a sensitivity analysis, we excluded the 17% (n=51 525) of the cohort who had been identified on the basis of internal contamination or monitoring. This exclusion had a much larger effect for the UK cohort than for those in the USA or France. The UK had identified anyone monitored for internal exposure, whereas the US and France identified anyone with a confirmed uptake.
In the present study, after the exclusion of workers flagged for internal contamination or monitoring, the estimated association between colon dose and mortality due to cancer other than leukaemia was larger in magnitude than the estimate for the whole cohort. This difference was consistent with a previous observation that among UK nuclear workers, radiation dose-cancer associations were smaller for workers who were potentially exposed to internal radiation than for those not exposed. After excluding cancers of the lung, liver, and bone, we observed that the magnitude of the estimated excess relative rate (0.51 per Gy; 90% confidence interval 0.15 to 0.91) was similar for all solid cancers. This estimate was larger than the estimated association between colon dose from external ionising radiation exposure and mortality due to solid cancers other than lung, liver, and bone among workers employed at the Mayak Production Association in Ozyorsk, Russia (0.16 per Gy; 0.08 to 0.24). Further work on internal doses is ongoing and could allow for increased attention to effects of incorporated radionuclides in future analyses.
Follow-up of large cohorts of nuclear industry workers has been ongoing for over 30 years; our data now yield sufficient statistical information to permit relatively precise estimates of cancer mortality risk in a population for whom average cumulative doses are about 20 mGy. These findings represent a substantial addition to the scientific basis for understanding the risks of cancer from protracted, low dose rate, exposure to ionising radiation; and underscore the value of the substantial efforts being made in France, the UK, and the USA to continue gathering data for these worker studies.
Discussion
Principal Findings
This study provides evidence of a linear increase in the excess relative rate of cancer mortality with increasing exposure to ionising radiation at the low dose rates typically encountered in the nuclear industries in France, the UK, and the USA. Restricting analyses to information regarding doses below 200, 150, and 100 mGy showed that the estimated excess relative rate per Gy for all cancers other than leukaemia were not driven by the highest dose categories. Analyses restricted to these lower doses also address the radiation protection community's interest in epidemiological evidence of a radiation dose-cancer association in these low dose ranges. INWORKS thus provides supportive evidence for a positive association between radiation dose and all cancer other than leukaemia, even if less precise when analyses are restricted to data for the 0–100 mGy dose range.
Comparison With Other Studies
The primary basis for the radiation risk estimates used to establish contemporary radiation protection standards comes from analyses of cancer in the Life Span Study of the Japanese atomic bomb survivors. Historically, it has been assumed that radiation-solid cancer associations diminish with falling dose rate. For example, the International Commission on Radiological Protection (2007) recommended that regulators divide the radiation risk coefficients obtained from the study of Japanese atomic bomb survivors in half when estimating risks for cancers other than leukaemia in settings with exposures of low dose and low dose rate radiation. This was recommended because the high dose rate exposures from the atomic bombings were assumed to be more dangerous than the low dose rate exposures typically encountered by workers and members of the public. Questions about the effects of such low dose exposure have, in part, motivated studies of nuclear workers since the 1970s.
Our estimated association between radiation and solid cancer (excess relative rate 0.47 per Gy; 90% confidence interval 0.18 to 0.79) is larger than but statistically compatible with the estimate from a mortality analysis of male survivors of the Japanese atomic bomb exposed at ages 20–60 years (excess relative rate 0.32 per Sv; 95% confidence interval 0.01 to 0.50). Statistical compatibility of risk estimates between studies may suggest some degree of coherence in the evidence derived from these large studies. However, in observational cohort studies, such as INWORKS and the Life Span Study of Japanese atomic bomb survivors, large sample sizes and statistical precision are no protection against bias. We have attempted to deal with some concerns regarding bias through decisions in study design. To this end, INWORKS was not intended to assemble the largest number of nuclear workers possible, but rather to assemble those cohorts that were most informative with regard to quality and completeness of exposure and follow-up data.
The parent study of the INWORKS collaboration included 407 391 workers from 15 countries. Although INWORKS included fewer nuclear workers than the earlier study, dose-response analyses in INWORKS encompassed substantially more cancer deaths than the parent study (17 957 v 4770 solid cancers), reflecting the extended follow-up of the INWORKS cohorts. INWORKS did not include data from Canada, a cohort for which the excess relative rate per Gy estimate was considerably larger than that observed in most other countries in the parent study, and for which concerns have been raised regarding data quality and completeness. In our analysis, no single country's data exerted a large impact on the magnitude of the summary risk estimate. Rather, a statistical test of heterogeneity by country rejected the conclusion of significant variation in the radiation-cancer association between the three countries.
The summary risk estimates in the current analysis are more precise and are larger in magnitude than those obtained from previous country specific analyses. For example, a previous analysis of data from France reported an estimated excess relative rate for solid cancers of 0.34 per Gy (90% confidence interval −0.56 to 1.38). Data from the UK National Registry for Radiation Workers had an estimated excess relative rate for death due to cancers other than leukaemia of 0.28 per Gy (0.02 to 0.56). Finally, data from the US showed an estimated excess relative rate for cancers other than leukaemia of 0.14 per Gy (−0.17 to 0.48).
Our estimated radiation risk coefficients could be somewhat larger than those in previous analyses of the constituent cohorts because we adjusted the recorded dose to account for bias in historical dosimeter response and attenuation, taking the estimated colon dose as the quantity of interest. Analysis of the INWORKS data using recorded photon dose as the dose metric, rather than adjusted estimates of colon dose, yielded somewhat lower estimates of association, although use of adjusted colon dose estimates resulted in no significant improvement in model goodness of fit (web Table A3). This finding accords with the general principle that it is possible to use assumptions about patterns of exposure misclassification (for example, different exposure periods or conditions) to reduce bias, but it is much more difficult to recover the precision that would be obtained if one knew each person's true exposure.
Strengths and Limitations of Study
Our adjusted dose estimates drew on the substantial work done to characterise the performance of the various radiation dosimeters used in France, the UK, and the USA over the study period and account for differences between countries and over time in dosimeter performance. Use of colon dose estimates facilitated comparison of our radiation risk estimates with those reported in mortality analyses of Japanese atomic bomb survivors that also related to estimated colon dose. However, exposure measurement errors related to personal dosimeters, monitoring practices, and historical records, particularly in the early years of operation, remain a study limitation. Radiation exposures might also have occurred outside of employment at facilities for which we have dosimetry records, and some workers could have had occupational radiation exposures that were not identified in the records available for this study.
In view of our focus on mortality due to cancer, a reasonable concern is potential confounding by cigarette smoking, which was unmeasured in our study. Contrary to the pattern that would be expected if there was confounding by smoking, the magnitude of the estimated excess relative rate per Gy under a 10 year lag was essentially unchanged after excluding lung cancer. A separate paper on non-cancer disease further supports the conclusion of no significant confounding by smoking as evidenced by the lack of association between radiation dose and chronic obstructive pulmonary disease, an outcome strongly associated with smoking. Although there has been interest in the joint effects of radiation and smoking on cancer risk, this could not be evaluated in our study.
Similarly, we could not directly adjust for the effects of exposure to other known occupational lung carcinogens, such as asbestos. However, the magnitude of the estimated association between radiation dose and solid cancer mortality remained unchanged after the exclusion of lung cancer (and further exclusion of pleural cancer). Therefore, occupational lung carcinogens may not be an important confounder in the overall analysis of the association between radiation and solid cancer.
In INWORKS, adjustment for socioeconomic status reduced the magnitude of dose-response estimates, suggesting positive confounding by socioeconomic status. This variable, which is primarily based on job title, is likely to be a poor proxy for factors that relate social class to mortality differences, and suggests the possibility of residual confounding of radiation-cancer associations by socioeconomic status. We adjusted for duration of work owing to evidence of a modest deficit in relative mortality among workers who had at least 10 years of radiation work, and a larger deficit in relative mortality among workers who had at least 30 years of radiation work. Stratification by duration of radiation work slightly increased estimates of association, suggesting negative confounding due to preferential retention of workers in better health (sometimes termed healthy worker survivor bias). We assessed the sensitivity to adjustment for these variables by fitting a simpler model that adjusted only for country, age, sex, and birth cohort. The estimated association between dose and cancer mortality other than leukaemia was similar in magnitude and precision to that obtained from the fully adjusted model, suggesting that the net effect of adjustment for these variables was small (web Table A4).
In the international collaborative study of cancer risk among radiation workers in the nuclear industry, people with potential exposure to neutrons, which was difficult to reliably quantify using the historical personnel dosimeters, were excluded from analyses to focus on workers with well-measured photon doses. A concern raised regarding this exclusion was that it excluded a large number of workers with high cumulative external photon doses.
In the current analysis, we included workers with potential exposure to neutrons and adjusted where possible for neutron monitoring status. In sensitivity analyses, we excluded the 13% of the cohort that ever had a recorded neutron dose. The resulting estimated association between colon dose and mortality due to cancer other than leukaemia (excess relative rate 0.55 per Gy; 90% confidence interval 0.17 to 0.95) was similar to that obtained for the whole cohort after adjustment for neutron monitoring status (0.48 per Gy; 0.20 to 0.79). However, owing to the limitations of historical neutron dosimetry information, an analysis restricted to workers with no recorded neutron dose is likely to have included workers who had unrecorded neutron exposures, particularly among those employed in the early years of operations.
The estimated association among workers with a positive recorded neutron dose (excess relative rate 0.36 per Gy; 90% confidence interval −0.08 to 0.88) and those with a recorded neutron dose exceeding 10% of total dose (0.62 per Gy; −0.50 to 2.09) were also similar in magnitude to, but less precise than, the whole cohort estimate after adjustment for neutron monitoring status. Therefore, the summary adjusted estimate does not appear to have obscured any meaningful heterogeneity in the excess relative rate per Gy by neutron monitoring status (web Table A5).
Employees who had recorded neutron doses—which reflects work in radiologically controlled areas where neutron dosimeters were issued—tended to have lower baseline rates of cancer mortality than those who did not. Reduced mortality rates among workers in radiologically controlled areas has been attributed to factors such as restrictions on smoking in such areas, and additional medical screening for work in areas where additional hazards might occur. Adjustment for neutron monitoring status accounts for such differences in baseline rates between groups, yielding a summary adjusted estimate comparable to the stratum specific estimates of excess relative rate per Gy. However, an unadjusted estimate produced a smaller value (0.20 per Gy; 90% confidence interval −0.03 to 0.46; web Table A5). Adjustment for neutron monitoring status, however, might be inadequate to fully control for differences in baseline rates between these groups, owing to limitations of historical neutron dosimetry information. Furthermore, bias could persist in adjusted analyses if health related selection out of employment affects a worker's future monitoring status and exposure history, and is itself affected by previous radiation exposure.
INWORKS included workers with potential for committed doses from incorporated radionuclides. As a sensitivity analysis, we excluded the 17% (n=51 525) of the cohort who had been identified on the basis of internal contamination or monitoring. This exclusion had a much larger effect for the UK cohort than for those in the USA or France. The UK had identified anyone monitored for internal exposure, whereas the US and France identified anyone with a confirmed uptake.
In the present study, after the exclusion of workers flagged for internal contamination or monitoring, the estimated association between colon dose and mortality due to cancer other than leukaemia was larger in magnitude than the estimate for the whole cohort. This difference was consistent with a previous observation that among UK nuclear workers, radiation dose-cancer associations were smaller for workers who were potentially exposed to internal radiation than for those not exposed. After excluding cancers of the lung, liver, and bone, we observed that the magnitude of the estimated excess relative rate (0.51 per Gy; 90% confidence interval 0.15 to 0.91) was similar for all solid cancers. This estimate was larger than the estimated association between colon dose from external ionising radiation exposure and mortality due to solid cancers other than lung, liver, and bone among workers employed at the Mayak Production Association in Ozyorsk, Russia (0.16 per Gy; 0.08 to 0.24). Further work on internal doses is ongoing and could allow for increased attention to effects of incorporated radionuclides in future analyses.
Conclusions and Implications for Future Research
Follow-up of large cohorts of nuclear industry workers has been ongoing for over 30 years; our data now yield sufficient statistical information to permit relatively precise estimates of cancer mortality risk in a population for whom average cumulative doses are about 20 mGy. These findings represent a substantial addition to the scientific basis for understanding the risks of cancer from protracted, low dose rate, exposure to ionising radiation; and underscore the value of the substantial efforts being made in France, the UK, and the USA to continue gathering data for these worker studies.
