Modeling time-to-tumor data: analysis of the ED01 study

Area of Science:

• Toxicology
• Carcinogenesis research
• Statistical modeling in biology

Background:

• The ED01 study involved exposing over 24,000 mice to 2-acetylaminofluorene, a known carcinogen.
• Discrepancies exist between the National Center for Toxicological Research (NCTR) analysis and the Society of Toxicology's ED01 task force findings.
• This research presents a separate analysis focused on accurately modeling the dose-time relationship in carcinogenic responses.

Purpose of the Study:

• To develop and apply a non-parametric approach for modeling carcinogenic responses.
• To evaluate the adequacy of factorable hazard function models in representing dose-time relationships.
• To compare the fit of different statistical models to liver and bladder neoplasm data.

Main Methods:

• Utilized a general non-parametric approach on data from sacrificed animals to avoid protocol complications.
• Assessed models with factorable hazard functions (separating dose and time effects).
• Compared model fits, including Weibull distributions and polynomial functions, against the Hartley-Sielken model.

Main Results:

• Factorable hazard function models are too simplistic for the entire experimental range of dose and time for both liver and bladder neoplasms.
• Excluding month 33 data improved the fit of factorable hazard models for liver neoplasms, showing a J-shaped dose response and power/polynomial time function.
• The Weibull distribution with dose affecting the scale parameter aligns with Druckrey's empirical equation for liver neoplasms.
• Bladder neoplasm data showed poor fit to factorable hazard models even after excluding high-dose/time points, indicating insensitivity to dose and time variations.

Conclusions:

• Simple factorable hazard models are inadequate for comprehensive carcinogenic response modeling.
• More complex models, like those incorporating Weibull distributions, are necessary for accurate representation of dose-time effects.
• The analysis highlights the challenges in modeling bladder neoplasms due to their limited sensitivity to dose and time variations within the observed range.