Modeling tumor onset and multiplicity using transition models with latent variables

Area of Science:

• Toxicology and Carcinogenesis
• Biostatistics
• Animal Modeling

Background:

• Transgenic rodent studies are increasingly used for carcinogen screening as an alternative to traditional chronic bioassays.
• Skin papilloma development in transgenic mice is a common endpoint, requiring methods to analyze tumor count data over time.

Purpose of the Study:

• To develop and present a statistical modeling method for analyzing carcinogenicity data from animal studies, focusing on tumor counts over time.
• To apply this method to transgenic rodent studies for evaluating potential carcinogens.

Main Methods:

• A statistical model is proposed assuming two unobservable latent variables per animal at each time point.
• The product of these latent variables follows a zero-inflated Poisson distribution.
• The Expectation-Maximization (EM) algorithm is used to maximize the observed-data pseudo-likelihood.
• Generalized estimating equations provide a robust variance estimator to account for within-animal outcome dependency.

Main Results:

• The developed method effectively models tumor counts over time in animal carcinogenicity studies.
• It allows for the assessment of dose-related trends in both tumor incidence and multiplicity.
• The approach is validated for application in transgenic rodent screening.

Conclusions:

• The proposed statistical modeling framework provides a powerful tool for analyzing complex tumor count data in carcinogenicity studies.
• This method enhances the utility of transgenic rodent models for carcinogen screening.
• It facilitates robust statistical inference on dose-response relationships in toxicological research.