A comparison of three methods for calculating confidence intervals for the benchmark dose

Area of Science:

• Toxicology
• Biostatistics
• Risk Assessment

Background:

• Calculating confidence intervals for benchmark dose (BMD) is crucial in risk analysis.
• Nonlinear dose-response models are frequently used in toxicology and risk assessment.
• Existing methods for BMD confidence intervals require careful evaluation for reliability.

Purpose of the Study:

• To compare the performance of three confidence interval calculation methods for benchmark dose estimation.
• To evaluate the reliability of the delta method, likelihood-ratio method, and bootstrap method in fitting nonlinear dose-response models.
• To identify the most suitable method for routine dose-response analysis and probabilistic risk assessment.

Main Methods:

• Comparison of the delta method, likelihood-ratio method, and bootstrap method.
• Fitting of nonlinear dose-response models with varying shapes to developmental toxicity data.
• Utilizing continuous, ordinal, and quantal dose-response data for method evaluation.

Main Results:

• The bootstrap method requires several thousand runs for stable confidence limits.
• The bootstrap and likelihood-ratio methods yield comparable results.
• The delta method produced different, often narrower, intervals and showed unreliability with nonlinear models.

Conclusions:

• The likelihood-ratio method is recommended for routine dose-response analysis due to its efficiency and reliability.
• The bootstrap method, while computationally intensive, offers advantages for direct integration with Monte Carlo analysis in probabilistic risk assessment.
• The delta method is not recommended for nonlinear dose-response models.