HAL Id: hal-02870744
https://hal.archives-ouvertes.fr/hal-02870744
Submitted on 6 Jul 2020
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Copyright
Estimation of lung cancer risk associated to multiple correlated sources of ionizing radiation in the post-55
French cohort of uranium miners
Marion Belloni, Chantal Guihenneuc, Sophie Ancelet
To cite this version:
Marion Belloni, Chantal Guihenneuc, Sophie Ancelet. Estimation of lung cancer risk associated to multiple correlated sources of ionizing radiation in the post-55 French cohort of uranium miners.
European Radiation Protection Week 2019, ERPW 2019, Oct 2019, STOCKHOLM, Sweden. 2019.
�hal-02870744�
Estimation of lung cancer risk associated to multiple correlated sources of ionizing radiation in the post-55 French cohort of uranium miners
Marion Belloni 1*, Chantal Guihenneuc 2, Sophie Ancelet 1
1 Ionizing Radiation Epidemiology Laboratory, IRSN; 2 EA7537, Faculté de Pharmacie de Paris, Université de Paris; * [email protected]
Context
Nuclear workers are chronically exposed to multiple sources of ionising radiation at low doses
Standard risk analysis Ô Single exposure scenario What about multiple exposure? Ô Few studies
Major issue Ô Exposure are potentially correlated Ô Re- gression models are not appropriated
Objective Ô Proposition of Bayesian hierarchical model to estimate radiation related risk from multiple correlated exposure
Case study
Aim : Estimating the risk of lung cancer in the post- 55 French cohort of uranium miners exposed to mul- tiple and correlated exposures to radon, gamma rays and uranium dusts
Miners who worked at least a year as a uranium miner in the company CEA-COGEMA and employed after 1955
Ý 3377 uranium miners Vital status n (%)
Alive 2486 (73.6)
Death from lung cancer 94 (2.8)
Death from another cause 777 (23.0)
Lost to follow-up 20 (0.6)
Mean (min-max) cumulated exposure:
Radon (WLM) 17.8 (0.01-128.4)
Gamma rays (mGy) 54.9 (0.20-470.1)
Uranium dusts (kBq.m´3.h) 1.64 (0.01-10.4)
Multicollinearity
If multiple regression
• Estimates have a larger variance
• Risk estimates are unstable
• Loss of statistical power
Posterior medians and 95% credible intervals for single / multiple exposure regression models:
Model βgamma/100mSv βradon/100WLM βdust/100Bq.h.m-3
Univariate 0.78 [ 0.28;1.67] 2.7 [ 1.1;5.2] 3.34e-2 [ 1.07e-2;7.00e-2]
Trivariate -0.00 [-0.39;1.17] 2.7 [-0.2;5.8] -0.15e-2 [-1.66e-2;3.81e-2]
Gamma Radon -0.03 [-0.35;1.06] 2.7 [ 0.0;5.6] .
Gamma Dust 0.72 [-0.12;1.78] . 0.54e-2 [-1.89e-2;5.42e-2]
Radon Dust . 2.7 [ 0.8;5.5] -0.17e-2 [-1.41e-2;3.47e-2]
• Single exposure : the associations are significant
• Multiple exposure : the associations are NOT significant
Our solution : Profile regression
Ô To identify clusters of uranium miners who have similar exposure profiles (ie similar caracteristics of exposure) and similar death risk
Ô To estimate the risk of death by lung cancer of the obtained clusters
The hierarchical model
Disease model : Survival model
• Survival data censored : Yi “ minpTi, Aiq, δiY “ 1Yi“Ti
y Ti : Age of miner i at death by lung cancer y Ai : Age of miner i at censoring
• Hazard rate by lung cancer : hiptq “ h0ptqp1 ` βCiq
y h0ptq : Baseline hazard rate (piecewise constant defined by λ) y βc : Excess hazard ratio of cluster c
Exposure model
Exposure profile Zi “ pZi,1, ..., Zi,P q of miner i with P characteristics
• Continuous variables : Zi,pCont|Ci “ c „ LogN ormalpµcp, σpcq
y Cumulative exposure to radon, gamma rays and uranium dusts lagged by 5 years of miner i
y Age of miner i at first exposure
• Discrete variables : Zi,pDisc|Ci “ c „ M ultinomialppcpq y Job type as a proxy for exposure conditions
y Mine where the miner mostly worked y Duration of exposure
Attribution model
• Unknown number of clusters
• Ci : cluster of miner i
• ψc : probability to be allocated to cluster c
• Vector of probabilities ψ modeled as "stick-breaking" prior with parameter α Ô The larger α, the larger the number of groups
Prior distributions
All weakly informative priors except λ and µ (for cumulative exposure)
Preliminary results from a Bayesian inference (
MCMC algorithm implemented in Python)8 clusters identified:
• 2 groups at significantly higher risks (G & H)
• 1 higher risk group at the limit of significance (F)
Gamma Radon
Cluster Dust Jobtype Hérault Age Duration
H +++ Before No . ě19 years
β « 1.4 mechanisation
G + After No Young 6-18 years
β « 1.2 mechanisation
F ++ Hewer & . . ě13 years
β « 0.5 Before
mechanisation
1
