Uncertainties in Fatal Cancer Risk Estimates Used in Radiation Protection

It is generally well known that the point estimates of the excess lifetime risk of fatal cancer following low dose/low dose rate exposure to ionising radiation are per Sv for a general population of all ages, and per Sv for adult workers. It is upon these two risk coefficients that radiation protection is principally founded. Until recently, less attention has been paid to the uncertainties associated with these risk estimates, although it is of some importance to have an appreciation of the degree of accuracy associated with them. In the US, the NCRP has recently addressed this issue through a Scientific Committee chaired by Warren Sinclair, with Andre Bouville and Charles Land as members. The result of their deliberations is to be found in NCRP Report No 126, which was published in October 1997. This important Report addresses the various sources of uncertainty present in the risk estimates which underlie radiological protection. As a consequence, attention is concentrated upon the overall fatal cancer risk arising from low dose/low dose rate, low LET radiation exposure, and other uncertainties, such as those associated with risk estimates for specific sites of cancer or with the internal deposition of alpha emitters, are not dealt with by this Report. The Committee focused upon cancer mortality estimates obtained from the Life Span Study (LSS) of the Japanese atomic bomb survivors and the application of these estimates to the conditions of radiological protection. In doing this, five broad areas of uncertainty were considered: epidemiological uncertainties, dosimetrical uncertainties, population transfer model, projection to lifetime and extrapolation to low dose or low dose rate exposure. For each of these five components of uncertainty, a likeliest value and a 90% subjective confidence interval (sometimes referred to as a 90% 'credibility interval') were derived and then the consequent overall uncertainty was estimated. The Report's treatment of epidemiological uncertainties covers random errors in observations and also systematic errors. Even in a large study such as the LSS, the number of excess deaths above expectation is limited, resulting in inevitable statistical uncertainty. The magnitude of this statistical uncertainty is estimated in the Report through a normal distribution having a standard deviation of 0.15. Ascertainment of cancer deaths is not without error; cases can fail to be detected or non-cancer cases recorded as cancers, leading to misclassification. The correction factor is taken to be normally distributed with a central value of 1.1 and a standard deviation of 0.05. Other possible sources of bias are that the wartime populations of Hiroshima and Nakasaki are unrepresentative of a general population and that differences exist between the populations of these two cities, but specific account is not taken of these possible factors in the analysis. Dosimetrical uncertainties arise from both random errors and biases in dose estimates. Random errors lead to an underestimation of the slope of the dose-response and a normal distribution for the required correction factor is derived with a modal value of 1.1 and standard deviation of 0.06. Biases in the dose estimates themselves arise from a number of sources: biases in both the gamma and neutron components of the dose, shielding uncertainties and uncertainty in the neutron RBE. When all the dosimetrical uncertainties are appropriately combined, the overall dosimetry uncertainty obtained is normally distributed with a most likely value of 0.84 and a standard deviation of 0.11. Having considered the uncertainties inherent within the risk estimates obtained directly from the LSS, the Report then moves on to examine the errors arising from applying them to the conditions relevant to radiological protection. The first set of these uncertainties analysed is that associated with the transfer of risk from the Japanese atomic bomb survivors to other populations that may have very different background rates of certain cancers. Usually, for a particular type of cancer, either the excess absolute risk or the relative risk is transferred, reflecting either an additive or a multiplicative model respectively. As the Report points out, these transfer models represent two extremes with the actual transfer mechanism lying somewhere in between, and a hybrid approach is discussed. It is fortunate that the differences between models for overall cancer mortality are less than the large uncertainties which can arise for specific types of cancer. For the US population, the additive and multiplicative methods of transfer give risk estimates which differ by about 30%. The distribution of uncertainties due to transfer of risk is taken to be a lognormal distribution with a most likely value of 1 and 5th and 95th percentile values of 0.70 and 1.65 respectively. Most of the Japanese atomic bomb survivors exposed at young ages are still alive and, therefore, to obtain lifetime risk estimates for this subcohort, projection from the observed time interval to lifetime risk is required. In deriving lifetime risk estimates, the conservative approach of assuming a constant relative risk over the remainder of life is taken, although evidence exists for a fall-off in the relative risk with either time since exposure or attained age. The uncertainty associated with a constant relative risk in lifetime projection is assumed to be triangularly distributed with a modal value of 1 and 5th and 95th percentile values of 0.62 and 1.05 respectively. The final area of uncertainty considered in detail in the NCRP Report is that due to extrapolation from the acute high exposure conditions experienced by the atomic bomb survivors to the chronic and/or low dose circumstances of radiological protection. This is achieved through the application of a dose and dose rate effectiveness factor (DDREF). There is considerable radiobiological evidence for a DDREF of greater than 1, but the epidemiological evidence is not so strong, particularly from the LSS for which a linear dose-response fits the solid tumour data well. The NCRP Scientific Committee concludes that the DDREF has a most likely value of 2 with a 5th percentile value of 1.25 and a 95th percentile value of 4.13. Finally, the report considers how all these various sources of uncertainty may be combined to produce an estimate of overall uncertainty. Particular uncertainties are not addressed (such as the possibly different effectiveness of irradiation by photons of a lower energy than those experienced from the bomb blast) and account is taken of these further identified sources of uncertainty, plus other unknown uncertainties, through an additional uncertainty factor. This factor is considered to be normally distributed with a central value of 1 and a standard deviation of 0.3. Unfortunately, the Report does not discuss the reasoning behind these values, which is surprising because this factor is influential in determining the overall uncertainty. I would like to have seen more discussion of this issue. When the various uncertainty factors are appropriately combined the mean value of the lifetime fatal cancer risk coefficient applicable to a US population of all ages is per Sv with a 90% subjective confidence interval from 1.2 to per Sv. This mean value is lower than the nominal value of per Sv recommended by ICRP. The greatest influence upon the uncertainty is due to the DDREF, accounting for around 38%, followed by unspecified uncertainties (29%) and population transfer (19%). The risk coefficient for an adult worker population is found to have a mean value of per Sv with 90% subjective confidence limits of 1.15 and per Sv. Again, the mean value is less than the ICRP nominal value of per Sv. Once more the DDERF has the greatest influence upon the overall uncertainty (40%), followed by unspecified uncertainties (31%) and population transfer (20%). This NCRP Report is a valuable contribution to a better understanding of the uncertainties involved in the risk estimates used for the purposes of radiological protection. I would recommend it to anyone who wishes to have a better acquaintance with the issues involved. Overall, the results are broadly reassuring: upper 90% subjective confidence limits are around a factor of 1.5 to 2.0 above nominal risk coefficients, and mean values of risk estimates are below these nominal values. The Report also shows what are the most influential factors affecting the uncertainty of risk estimates, and, therefore, where most effort needs to be directed to reduce overall uncertainties. One difficult area which does need more attention is the treatment of unspecified uncertainties, a factor which, as can be seen from the above, does exert a significant influence on the analysis carried out by the NCRP Scientific Committee. This should not detract, however, from the substantive treatment of uncertainties in the risk estimates used for the purposes of radiological protection contained within NCRP Report No 126.