Estimation of lung cancer risk associated to multiple correlated sources of ionizing radiation in the post-55 French cohort of uranium miners

In radiation epidemiology, the health effects of occupational exposures are often studied separately for each source of ionizing radiation (IR). Radiation protection standards are thus mainly based on risk analyses focusing on a single exposure. But, nuclear workers are exposed to several sources of correlated exposures such as IR, chemical and physical agents. In this work, we focus on the risk of death by lung cancer in the post-55 French cohort of uranium miners who are chronically exposed to low doses of radon, gamma rays and uranium dust. We propose a Bayesian hierarchical approach called "Bayesian Profile Regression". This model makes it possible to estimate the radiation-related risk of death by lung cancer in the specific context of multiple and correlated exposures to IR, for which simple multiple regressions may lead to unstable and unprecise risk estimates. It is based on the combination of three conditionally independent sub-models: a survival disease sub-model, a classification sub-model and an exposure sub-model. Fitting this model under the Bayesian paradigm allows clustering individuals with similar profiles, that is, with similar exposure characteristics, and estimating the associated excess hazard risk (EHR) of radiation-related death by lung cancer for each group, in a unique step. Finally, the obtained results are post-processed to identify and characterize profiles of uranium miners with high or low EHR. Our model distinguished eight different profiles of uranium miners corresponding to eight different EHR. Among them, two profiles were associated with a strictly positive and statistically significant EHR. The first group (EHR=1.4, 95%IC=[0.60, 2.60]) corresponded to the miners the most highly exposed to radon, gamma rays and uranium dust and for more than 19 years (mainly before mechanization). The second group (EHR=1.2, 95%IC=[0.17, 2.80]) corresponded to miners who were very young when first exposed and exposed at high doses to radon, gamma rays and uranium dust (mainly after mechanization). Finally, the model showed that the group of miners who worked mainly in the Herault mine- the only included uranium mine with sedimentary soil- had lower EHR. These results do not account for the smoking status of miners for whom information is strongly missing in the cohort. Our short term perspectives are thus to try to account for the smoking status in our Bayesian profile regression but also on exposure measurement errors on radon, gamma rays and uranium dust.