Zero-inflated regression models for radiation-induced chromosome aberration data: A comparative study.

Area of Science:

• Cytogenetic biodosimetry
• Statistical modeling
• Radiation biology

Background:

• Classical Poisson regression is used for modeling chromosome aberrations in cytogenetic biodosimetry.
• Data often exhibit overdispersion, violating the equidispersion assumption of Poisson models.
• Unobserved heterogeneity in cell populations can lead to increased variance and zero counts.

Purpose of the Study:

• To compare different methodologies for analyzing cytogenetic chromosomal aberrations datasets.
• To focus on the application of zero-inflated Poisson and zero-inflated negative binomial models.
• To develop a score test for detecting zero inflation in Poisson regression models.

Main Methods:

• Comparison of statistical models for cytogenetic biodosimetry data.
• Application of zero-inflated Poisson (ZIP) models.
• Application of zero-inflated negative binomial (ZINB) models.
• Development of a score test for zero inflation in Poisson regression.

Main Results:

• Overdispersion is a common issue in cytogenetic biodosimetry data, particularly with partial-body radiation exposure.
• Zero-inflated models provide a more appropriate framework for analyzing such data compared to standard Poisson regression.
• A novel score test for zero inflation in Poisson models was developed.

Conclusions:

• Standard Poisson regression is often inadequate for analyzing chromosome aberration data due to overdispersion.
• Zero-inflated models, such as ZIP and ZINB, offer improved accuracy in modeling radiation dose effects.
• The developed score test aids in identifying and addressing zero inflation in cytogenetic biodosimetry studies.