Robust analysis of a mixed-effect model for a multicenter clinical trial

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Robustness

Background:

• Multicenter clinical trials often involve complex random effects (centers, residuals).
• Standard statistical methods may fail when normality assumptions for random effects are violated.
• Robustness to non-normality is crucial for reliable clinical trial analysis.

Purpose of the Study:

• To propose and evaluate statistical methods robust to non-normality in multicenter trials.
• To compare the performance of weighted least-squares (WLS) and bootstrap confidence intervals for treatment differences.
• To introduce a bootstrap method specifically designed to handle outliers.

Main Methods:

• Utilized weighted least-squares (WLS) for fixed effects, accommodating non-normal random effects.
• Implemented percentile and BCa bootstrap methods for robust confidence interval estimation.
• Conducted simulation studies to compare WLS and bootstrap interval performance.

Main Results:

• All evaluated methods (WLS, percentile bootstrap, BCa bootstrap) achieved desired confidence interval coverage rates.
• The WLS method consistently produced shorter confidence intervals compared to bootstrap methods.
• A novel bootstrap method demonstrated robustness against outliers.

Conclusions:

• WLS and bootstrap methods offer robust alternatives to conventional approaches in multicenter trials with non-normal random effects.
• WLS provides a more efficient estimation of treatment differences, yielding shorter intervals.
• The proposed outlier-robust bootstrap method enhances reliability in the presence of data anomalies.