Radiation biology. Radioecology
ISSN: 0869-8031

Archives → 2022 → Vol. 62, № 3 → pp. 273–286

Article

Conditional Lifetime Radiation Riskin a Heterogeneous Hibakusha Cohort

Obesnyuk V. F.

Southern Urals Biophysics Institute FMBA, Ozyorsk, Russia

Abstract

A new design of a retrospective group analysis of the dependence of the oncological death risk on the simultaneous action of a known influencing factors set, including ionizing radiation, is proposed. The conditional lifetime risk is chosen as the studied rate in contrast to the broad international practice, when the annual intensity rates of specific events are used for observation. The efficiency of the research method was tested on the example of the published death register of Hiroshima and Nagasaki residents, who survived after the atomic bombing. The central idea of the applied method of analysis was the rejection of traditional heuristic single-factor models of the risk rates in favor of the use of multi-factor artificial neural networks as an automatic generator of risk models. The achieved approximation quality turned out to be higher, which allowed us to identify any significant trends that were previously statistically indistinguishable. According to the indicator of lifetime risk rate, the «dose–risk» trend deviates significantly from a linear relationship in a heterogeneous cohort of people who survived the atomic bombing. The linear law with a single dose coefficient of nominal risk can be justified only if the cohort paradigm is abandoned in favor of population estimates in the range of relatively small doses (up to about 1000 mSv) due to averaging by sex and age at exposure. The main reason is the influence of competing non-oncological causes of death with increasing dose. Based on the observation of abnormally high spontaneous risks of cancer mortality, it is assumed that not all active risk factors are taken into account in the published Japanese register. This may be the reason for the mathematical bias of trend estimates. The conclusion is based on the observation of abnormally high spontaneous risks of cancer mortality.

Keywords

Risk, rate, epidemiology, risk factor, competing risks, indirect estimation, mortality, neural network, cohort, stratum, model