Bayesian dose-response analysis for epidemiological studies with complex uncertainty in dose estimation

摘要

Most conventional risk analysis methods rely on a single best estimate ofexposure per person which does not allow for adjustment for exposure-relateduncertainty. Here, we propose a Bayesian model averaging method to properlyquantify the relationship between radiation dose and disease outcomes byaccounting for shared and unshared uncertainty in estimated dose. Our Bayesianrisk analysis method utilizes multiple realizations of sets (vectors) of dosesgenerated by a two-dimensional Monte Carlo simulation method that properlyseparates shared and unshared errors in dose estimation. The exposure modelused in this work is taken from a study of the risk of thyroid nodules among acohort of 2,376 subjects following exposure to fallout resulting from nucleartesting in Kazakhstan. We assessed the performance of our method through anextensive series of simulation tests and comparisons against conventionalregression risk analysis methods. We conclude that when estimated doses containrelatively small amounts of uncertainty, the Bayesian method using multiplerealizations of possibly true dose vectors gave similar results to theconventional regression-based methods of dose-response analysis. However, whenlarge and complex mixtures of shared and unshared uncertainties are present,the Bayesian method using multiple dose vectors had significantly lowerrelative bias than conventional regression-based risk analysis methods as wellas a markedly increased capability to include the pre-established 'true' riskcoefficient within the credible interval of the Bayesian-based risk estimate.An evaluation of the dose-response using our method is presented for anepidemiological study of thyroid disease following radiation exposure.