Integrating informative priors from experimental research with Bayesian methods: an example from radiation epidemiology

*Ghassan Hamra, University of North Carolina at Chapel Hill 
Richard MacLehose, University of Minnesota 
David Richardson, University of North Carolina at Chapel Hill 
Steve Wing, University of North Carolina at Chapel Hill 

Keywords:

The relationship between exposure to tritium and risk of cancer has received little attention in the epidemiological literature. This may be due to challenges in risk estimation due to sparse data. For radiation protection purposes, and for radiation risk assessments, tritium dose and gamma radiation doses are often combined, assuming that the relative biological effectiveness of each is equal. Evidence from experimental animal and cellular literature shows that tritium is more biologically damaging than gamma radiation. However, integrating this information into epidemiological research is challenging. We present a method to help bridge this gap between animal and cellular studies and epidemiological research by specification of an order constrained prior. We utilize Bayesian methods in order to evaluate the excess relative rate (ERR) of leukemia and leukemia excluding CLL per unit of absorbed dose of tritium informed by its relative effect compared to gamma radiation. The ERR/10mGy and associated 90% highest posterior density (HPD) of leukemia and leukemia excluding CLL are 0.272 (0.024, 0.660) and 0.304 (0.044, 0.753). This is the first empirical estimate of tritium’s relationship to cancer risk based on epidemiological data.