Robust designs accounting for model uncertainty in longitudinal studies with binary outcomes.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Longitudinal Data Analysis

Background:

• Optimal design of longitudinal studies often relies on a priori model knowledge, limiting flexibility.
• Mixed-effect models are commonly used for analyzing longitudinal data with binary outcomes.

Purpose of the Study:

• To propose a novel robust design approach for longitudinal trials with two treatment groups, specifically addressing model uncertainty.
• To enhance the precision of parameter estimation and the power of statistical tests in the presence of model uncertainty.

Main Methods:

• Utilizing the Fisher information matrix and Monte-Carlo/Hamiltonian Monte-Carlo methods for design optimization.
• Employing the DD S -optimality criterion to balance parameter estimation precision and Wald test power.
• Incorporating model uncertainty by considering weighted candidate models and computing a compound DD S -optimality.

Main Results:

• The robust design approach demonstrated efficiency across various models in simulations.
• The proposed method successfully achieved good statistical power for hypothesis testing.
• Predicted average power using the Fisher information matrix aligned with simulation outcomes.

Conclusions:

• The developed robust design strategy offers a new and valuable method for planning longitudinal studies with binary outcomes.
• This approach effectively manages model uncertainty, leading to more reliable trial designs.
• The findings support the use of this robust design in optimizing longitudinal clinical trials.