Logistic Regression Models with Unspecified Low Dose-Response Relationships and Experimental Designs for Hormesis...

Area of Science:

• Toxicology
• Environmental Science
• Biostatistics

Background:

• Hormesis describes nonmonotonic (biphasic) dose-response relationships.
• Low toxicant doses may pose less risk than control doses, with risk increasing at higher doses.
• Small sample sizes necessitate efficient parametric models for hypothesis testing, despite strong assumptions.

Purpose of the Study:

• To develop robust parametric models for hormesis research.
• To evaluate alternative logistic regression parameterizations and experimental designs.
• To improve hypothesis testing accuracy in the presence of model misspecification.

Main Methods:

• Considered alternative parameterizations of the traditional three-parameter logistic regression.
• Investigated uniform and D-optimal experimental designs.
• Conducted simulation studies to assess Type I error rates and statistical power.

Main Results:

• D-optimal design with traditional logistic regression failed to control inflated Type I error rates due to model misspecification.
• New three-parameter logistic regression parameterizations with D-optimal design maintained Type I error rates near the significance level.
• New four-parameter parameterization with D-optimal design preserved Type I error rates with minimal power reduction.

Conclusions:

• Alternative parameterizations enhance the robustness of logistic regression models for hormesis.
• Optimal experimental designs combined with appropriate parameterizations improve hypothesis testing reliability.
• The proposed methods offer a more accurate approach to analyzing hormesis data, especially with limited sample sizes.