Bayesian hierarchical modeling to account for complex patterns of measurement error in cohort studies. Application in radiation epidemiology.

International audienceDespite its deleterious consequences for statistical inference and its ubiquity in observationalresearch, exposure measurement error is rarely accounted for in epidemiological studies. Basically,standard correction methods, like regression calibration or SIMEX, often lack the flexibility to accountfor complex patterns of exposure uncertainty. However, in occupational cohort studies, for instance,changes in the methods of exposure assessment can lead to complex error structures. Moreover, astrategy of group-exposure assessment and individual worker characteristics may lead to errorcomponents that are shared between or within individuals. In this work, we thus propose and fit severalBayesian hierarchical models, combining survival models with time-dependent covariates,measurement and exposure models, to obtain a corrected estimate ofthe potential association betweenexposure to radon and mortality for several cancer types in the French cohort of uranium miners. Asimulation study is under progress to assess the impact of model misspecification on risk estimates