A major source of uncertainty in radiation risk assessments concerns extrapolation of risks from high to low dose. Crucial to the resolution of this uncertainty is the flexible modelling of the dose response relationship and the importance of shared vs unshared dosimetric errors. It is well known that measurement error can alter substantially the shape of this relationship and hence derived risk estimates.
Regression calibration (RC) is the simplest method of dose-error correction, and works well when errors are modest. When errors are large or shared, full-likelihood methods, in particular Markov Chain maximum likelihood (MCML) and Bayesian Markov Chain Monte Carlo (MCMC), may work better.
Here results of analyses of cancer data in the Japanese atomic bomb survivors and in groups in Ukraine exposed following the 1986 Chernobyl accident are analyzed, adjusting for dose error via a number of different methods (in particular RC, MCML and MCMC).
RC yields comparable risk estimates to the full likelihood methods, and all methods of dose-error adjustment have comparatively modest effects, a consequence of the relatively small errors; for the Ukraine data the errors are a mixture of Berkson and classical form.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.