Model selection and averaging of nonlinear mixed-effect models for robust phase III dose selection

Area of Science:

• Pharmacometrics
• Clinical Trial Design
• Drug Development

Background:

• Population modeling (pharmacometrics) is crucial for accurate dose selection in clinical trials.
• Reliance on a single model can introduce bias, compromising dose selection for subsequent trial phases.
• Model selection bias is a significant concern in phase IIb trial data analysis.

Purpose of the Study:

• To introduce and evaluate methods for combining or selecting candidate models to address model uncertainty.
• To improve the robustness of dose selection decisions at the end of phase IIb trials.
• To mitigate model selection bias in population model-based analyses.

Main Methods:

• Investigated four methods combining or selecting pre-defined dose-response model structures.
• Utilized realistic simulation studies based on an actual phase IIb clinical trial protocol.
• Employed a bootstrap model selection approach to assess its performance.

Main Results:

• The bootstrap model selection method was shown to effectively avoid model selection bias.
• This method generally increased the accuracy of decisions made at the end of phase IIb trials.
• Simulation studies confirmed the benefits of addressing both model structure and parameter uncertainty.

Conclusions:

• The bootstrap model selection method is recommended for population model-based decision-making in phase IIb trials.
• This approach enhances the reliability of dose selection by accounting for model uncertainty.
• Robust dose selection is vital for successful progression to phase III clinical trials.