Monotonic Bayesian semiparametric benchmark dose analysis.

Area of Science:

• Toxicology
• Biostatistics
• Risk Assessment

Background:

• Quantitative risk assessment relies on dose-response models, where model choice introduces significant uncertainty.
• Current methods for handling model uncertainty may fail when the true model is outside the considered space.
• Estimating the dose for a specific response level is crucial for risk assessment.

Purpose of the Study:

• To propose a novel semiparametric dose-response model that only assumes monotonicity.
• To derive a dose estimate associated with a prespecified response level using this model.
• To compare the proposed method with existing model uncertainty approaches using a cancer bioassay and simulations.

Main Methods:

• Developed a semiparametric dose-response model with a monotonicity constraint.
• Applied the model to estimate dose-response curves from a long-term cancer bioassay.
• Conducted a simulation study to compare performance against model averaging.

Main Results:

• The proposed semiparametric model accurately estimates dose-response relationships.
• The method outperforms model averaging for quantal-linear dose-response mechanisms.
• Performance is comparable to model averaging for nonlinear dose-response patterns.

Conclusions:

• The novel semiparametric model provides a more reliable approach to quantitative risk assessment.
• This method effectively addresses model uncertainty, particularly in toxicological studies.
• The approach enhances the estimation of safe exposure levels in risk evaluations.