CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0284
Title: Quasi-Poisson regression models for radiation dose estimation from biomarkers Authors:  Jochen Einbeck - Durham University (United Kingdom) [presenting]
Abstract: Poisson regression models have a long tradition in the construction of dose-response calibration curves from count-data valued biomarkers. For instance, the current `gold-standard' in radiation biodosimetry, based on dicentric chromosomes, usually adheres well to the equidispersion assumption of the Poisson distribution. However, there do exist several alternative modern biomarkers which allow for considerably quicker and cheaper analysis than the dicentric one, but often these come at the price of considerable overdispersion, for instance due to inter-individual variation. This holds particularly for the gamma-H2AX assay, a protein-based biomarker which makes use of counts of fluorescent dots produced by the H2AX histone following radiation-induced double-strand breaks. We illustrate the quasi-Poisson model in the context of the gamma-H2AX assay, and show how it can be used to quantify the uncertainty of doses estimated through this biomarker. Finally, the possibility of applying this procedure onto certain cytogenetic biomarkers which feature considerable overdispersion, such as micronuclei, is briefly discussed.