Robust and Adaptive Two-stage Designs in Nonlinear Mixed Effect Models.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Nonlinear mixed effect models (NLMEM) are crucial for analyzing complex longitudinal data.
• Optimizing study designs, particularly in clinical trials, requires accurate models and parameters, often necessitating adaptive strategies.
• Existing methods for design optimization, like D-criterion based on Fisher Information Matrix (FIM), rely on prior parameter guesses, limiting their robustness.

Purpose of the Study:

• To propose a novel two-stage adaptive design strategy for NLMEM.
• To enhance design optimization using a robust expected FIM and model averaging (MA) over candidate models.
• To evaluate the proposed strategy in a simulated ophthalmology clinical trial for optimizing dose and timing.

Main Methods:

• A two-stage adaptive design framework was developed, incorporating robust expected FIM and MA.
• Candidate dose-response models were defined for simulation.
• One-stage and two-stage (50/50 split) designs were compared using local optimal, robust designs, and employing single-model analysis, model selection (MS), or MA.

Main Results:

• The two-stage adaptive design with MS at interim analysis demonstrated the ability to correct initial model selection errors.
• Robust designs (both one-stage and two-stage) proved valuable, yielding acceptable bias and precision.
• The proposed robust adaptive design strategy offers improved performance over traditional methods.

Conclusions:

• The developed robust adaptive design strategy is effective for optimizing longitudinal studies, particularly in clinical settings.
• This approach provides a valuable tool for designing studies where model uncertainty is present.
• The methodology is adaptable for various therapeutic areas beyond ophthalmology.