A comparison of three methods for integrating historical information for Bayesian model averaged benchmark dose...

Area of Science:

• Environmental Health
• Toxicology
• Biostatistics

Background:

• The benchmark dose (BMD) approach is crucial for risk assessment but struggles with integrating diverse environmental hazard data.
• Selecting appropriate BMD/BMDL estimates from multiple dose-response models remains a challenge.

Purpose of the Study:

• To investigate the impact of historical, particularly incompatible, data on Bayesian model-averaged BMD estimates using various integration methods.
• To compare pooled data analysis, Bayesian hierarchical models, and the power prior method for integrating historical information.

Main Methods:

• Bayesian model averaging (BMA) was employed for benchmark dose estimation.
• Three integration methods were examined: pooled data analysis, Bayesian hierarchical models, and the power prior method.
• Dose-response models were analyzed to assess parameter distributions and uncertainty.

Main Results:

• Pooled data analysis significantly impacts BMA BMD estimates but has limited applicability and potential flaws.
• Bayesian hierarchical models moderately influence current estimates by accounting for between- and within-study uncertainty.
• The power prior method shows minimal impact on current estimates with highly incompatible data, even with full prior consideration.

Conclusions:

• The method of integrating historical data substantially affects model weights and BMD estimates.
• Bayesian hierarchical models provide a structured approach for combining information with reasonable impact.
• Careful consideration of data compatibility and integration methods is essential for robust BMD estimation in risk assessment.