Valid versus invalid radiation cancer risk assessment methods illustrated using Swiss population data

Abstract

After the nuclear accident in Fukushima, the public interest in radiation related cancer-risk assessment increased. However, interpretations of results from epidemiological studies and comprehension of cancer risk assessment methods can be unclear and involve questions about correctness and validity of the approaches. To shed some light on this potential lack of clarity, valid versus invalid radiation cancer risk assessments methods are illustrated here using Swiss population data. This involves a comparison of the cancer risk assessment method based on collective dose and the cumulative risk assessment method, where the latter is recommended with regard to uncertainties and risk of misinterpretation. Further, risk assessment in different dose ranges is discussed and it is concluded that below 100 mSv it cannot be appropriately stated that an adequate strength of evidence of a causal relationship between cancer and radiation is provided, because of the large uncertainties in this dose range. However, the linear non-threshold (LNT) model can be used to model the dose response, because it represents a prudent and parsimonious model, that fits the data well and lies within the given uncertainties. Additionally, treatments of uncertainties in the risk models are illustrated. The EU-project CONFIDENCE software is applied here to obtain example radiation related lifetime cancer risks for exposures of 20 mSv and 5 mSv. Furthermore, the impact of different dosimetry errors on the uncertainties in the cancer lifetime risk calculation is analysed, by including different standard deviations (SD) and by comparing the sampling of the doses from a normal and a lognormal distribution. Using the normal distribution, for females exposed to 20 mSv, the 95% confidence interval (CI) on the cancer lifetime risk increases, when compared to using a SD of 4 mSv, by a factor of 1.5 using a SD of 8 mSv and by a factor of 1.7 using a SD of 10 mSv. The corresponding factors for males for the same exposure are 1.3 and 1.5 respectively. For exposure to 5 mSv, the 95% CIs on the risk increase by a factor of 1.2 for females and 1.4 for men for a SD of 2 mSv using the normal distribution compared to the lognormal distribution and by a factor of 1.5 and 1.8 for a SD of 3 mSv compared to a SD of 1 mSv respectively. Furthermore, differences in the resulting 95% CI on the risk, using different distributions for the dose sampling are visible.