Statistics Seminars: Statistical models for radiation biodosimetry -- Poisson or not Poisson?
23 February 2015 14:00 in CM221
After the occurrence of a radiation accident or incident leading to irradiated blood lymphocytes, there is need for rapid and reliable procedures to determine the radiation dose contracted by the individuals. Since members of the public do not usually wear dosimeters, there is need for techniques which exploit the radiation-induced change in certain biomarkers to estimate the radiation dose directly for individuals, to inform triage and clinical decision making. This field is known as biological dosimetry, with the majority of methods using cytogenetic biomarkers such as dicentric chromosome aberrations or micronuclei in blood samples.
The Poisson distribution is the widely accepted count distribution for modelling the number of chromosome aberrations in blood samples. However, very often this distribution is in fact not Poisson, especially for densely ionising radiation and for partial body exposure, so that alternative models, such as zero-inflated models, should be used for this purpose. Score tests can be used to decide between such models, and we develop a new score test which can be used under the identity link function (which is is the link to be used according to the IAEA manual). We finish with some words on current work using alternative biomarkers (microarrays, H2AX).
Contact firstname.lastname@example.org for more information