Bayesian analysis of overdispersed chromosome aberration data with the negative binomial model

Area of Science:

• Radiation Biology
• Statistical Modeling
• Dosimetry

Background:

• Chromosome aberration data often assumes a Poisson model where variance equals the mean.
• High linear energy transfer (LET) radiation experiments frequently yield overdispersed data, meaning variance exceeds the mean.
• Current methods for handling overdispersed chromosome data primarily use frequentist statistics.

Purpose of the Study:

• To address overdispersion in chromosome aberration data using a Bayesian statistical framework.
• To compare Poisson and negative binomial models for analyzing dicentric chromosome aberrations induced by neutrons.
• To evaluate the impact of different model assumptions on neutron dose estimation.

Main Methods:

• Bayesian statistical analysis was employed to model chromosome aberration data.
• The Bayes Factor was utilized to compare Poisson and negative binomial models.
• Posterior densities for model parameters, including dose response and overdispersion, were calculated and visualized.
• Calibrative densities were derived for estimating unknown neutron doses.

Main Results:

• The study analyzed two published calibration datasets of dicentric chromosome aberrations induced by high neutron doses.
• Posterior densities revealed characteristics of dose response and overdispersion.
• Dose estimates were calculated, highlighting the influence of model assumptions.

Conclusions:

• A Bayesian approach provides a robust method for analyzing overdispersed chromosome aberration data.
• The negative binomial model is recommended over the Poisson model for high LET radiation dosimetry.
• Assuming a negative binomial model initially is a conservative strategy for accurate chromosome dosimetry in radiation accidents.