Model Uncertainty and Bayesian Model Averaged Benchmark Dose Estimation for Continuous Data.

Area of Science:

• Toxicology and Risk Assessment
• Statistical Modeling
• Environmental Health

Background:

• The benchmark dose (BMD) approach is crucial for risk assessment but faces challenges in selecting appropriate model estimates.
• Current methods often neglect explicit consideration of model uncertainty, limiting comprehensive risk assessment.
• There is a need for improved methods to inform health risk assessors about model uncertainty.

Purpose of the Study:

• To propose and evaluate a Bayesian model averaging (BMA) method for estimating benchmark dose (BMD) with continuous data.
• To address the challenge of model uncertainty in BMD estimation for risk assessment.
• To provide a more reliable alternative to current BMD estimation approaches.

Main Methods:

• Applied two BMA strategies (maximum likelihood estimation-based and Markov Chain Monte Carlo-based) using the Crump "hybrid" method.
• Calculated model-averaged BMD estimates from real continuous dose-response data.
• Conducted a simulation study to evaluate the accuracy and bias of the BMA BMD estimator.

Main Results:

• BMA BMD estimates demonstrated higher reliability, indicated by higher BMDL values and smaller 90th percentile intervals compared to individual models.
• Simulation results showed BMA BMD estimates have smaller bias than BMDs selected by other criteria.
• The BMA method offers a more robust approach to handling model uncertainty in BMD estimation.

Conclusions:

• The proposed BMA method effectively incorporates model uncertainty into BMD estimation for continuous data.
• BMA provides more reliable and less biased BMD estimates, enhancing risk assessment accuracy.
• Further research is needed on model selection and bootstrap methods for BMDL derivation within the BMA framework.