A Bayesian hierarchical approach to account for shared exposure measurement error in an occupational cohort - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

A Bayesian hierarchical approach to account for shared exposure measurement error in an occupational cohort

Résumé

Exposure measurement error poses one of the most important threats to the validity of statistical inference in epidemiological studies. Indeed, when exposure measurement is not or only poorly accounted for, it may lead to biased risk estimates, loss in statistical power and distortions of the exposure-response relationship. Despite these deleterious consequences and despite its ubiquity in observational research, exposure measurement is rarely accounted for in the estimation of risk coefficients in epidemiological studies. This may partly be due to the fact that classical methods that are routinely used to correct for measurement error, like simulation extrapolation or regression calibration, lack the flexibility to account for complex patterns of exposure uncertainty. In occupational cohort studies, for instance, changes in the methods of exposure assessment can lead to complex structures of exposure measurement error. Moreover, the use of a strategy of group-exposure assessment, for instance via job exposure matrices, and individual worker characteristics may lead to error components that are shared between or within workers, respectively. We propose several hierarchical models and conduct Bayesian inference for these models to obtain corrected risk estimates on the association between exposure to radon and its decay products and lung cancer mortality in the French cohort of uranium miners. The hierarchical approach, which is based on the combination of sub-models that are linked via conditional independence assumptions, provides a flexible and coherent framework for the modelling of complex error structures. We observe a marked increase in the excess hazard ratio when accounting for shared measurement error in a proportional hazards model whereas the correction for unshared measurement error is only of marginal importance in risk estimation. These results, which are in accordance with previous results on the impact of different measurement error characteristics in the French cohort of uranium miners obtained on simulated data, underline the importance of a careful characterization of all components of exposure measurement error in an occupational cohort study. In this context, the use of a Bayesian hierarchical approach provides the possibility to integrate expert knowledge or to combine epidemiological data with experimental laboratory data in a coherent framework.
0000161601_001.PDF (5.95 Mo) Télécharger le fichier
Format Présentation
Origine Fichiers produits par l'(les) auteur(s)
Licence
Copyright (Tous droits réservés)

Dates et versions

hal-02635580 , version 1 (19-08-2020)

Licence

Copyright (Tous droits réservés)

Identifiants

  • HAL Id : hal-02635580 , version 1

Citer

Sabine Hoffmann, Chantal Guihenneuc, Sophie Ancelet. A Bayesian hierarchical approach to account for shared exposure measurement error in an occupational cohort. 40th Annual Conference of the International Society for Clinical Biostatistics, ISCB, Jul 2019, LOUVAIN, Belgium. ⟨hal-02635580⟩
24 Consultations
9 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More