Dose–responses from multi-model inference for radiation-induced cardiovascular diseases in atomic bomb survivors.

This presentation focuses on two important detrimental health outcomes from the atomic bomb survivors of the Life Span Study (LSS; Preston et al. Radiat. Res. 2003, Shimizu et al. BMJ 2010). These two endpoints are mortalities from cerebrovascular diseases (CVD) and cardiovascular diseases excluding CVD (sometimes referred to as heart diseases). The data were analyzed with a larger set of plausible empirical models such as linear-no-threshold and quadratic models and a number of threshold, step and U-shaped models. They were combined with a suitable baseline model either as excess relative risk or as excess absolute risk models. The likelihood-ratio test was used to eliminate nested models. That way for each of the two detrimental health outcomes one sub-set of non-nested models was identified that all fitted the data about equally well. The two sub-sets of non-nested models were then used to perform multi-model inference, an innovative method of mathematically combining different models to allow risk estimates to be based on several plausible dose–response models rather than just relying on a single model of choice. Our findings for the LSS Report 13 data (Preston et al. 2003) support nonlinear dose-responses for CVD and cardiovascular diseases excluding CVD. The presentation will provide an update on our ongoing analyses of the latest LSS data that contain seven additional years of follow-up time (Shimizu et al. 2010). In opposition to the earlier study (Schöllnberger et al. REB 2012) the latest analyses apply the full data sets with follow-up starting in 1950 and including proximal and distal survivors in line with comments by Little et al. (REB 2013)