Adaptive trial design: a general methodology for censored time to event data

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Survival Analysis

Background:

• Adaptive clinical trial designs permit modifications based on interim data without compromising statistical integrity (type I error).
• The Inverse Normal method offers an adaptive generalization of group sequential designs.
• Prior applications of the Inverse Normal method for censored survival data were limited to the logrank statistic, which is suboptimal with nuisance covariates.

Purpose of the Study:

• To demonstrate the application of the Inverse Normal method with Cox regression analysis for adaptive clinical trials.
• To address the inefficiency of the logrank statistic in the presence of nuisance covariates.

Main Methods:

• The study extends the Inverse Normal method to Cox regression analysis for adaptive trial designs.
• Two approaches are presented to ensure the independence of test statistics across trial stages: utilizing the score process's independent increment structure and employing censoring/truncation to segment follow-up data.
• Simulation studies were conducted to evaluate the performance of these methods.

Main Results:

• The performance of the adaptive design was found to be independent of the specific method used to achieve statistical independence between stages.
• Adaptive Cox regression analysis demonstrated superior efficiency compared to adaptive logrank analysis when nuisance covariates influenced survival outcomes.

Conclusions:

• The Inverse Normal method can be effectively applied to Cox regression for adaptive clinical trials involving censored survival data.
• This adaptive Cox regression approach offers enhanced statistical efficiency, particularly when nuisance covariates are present, outperforming traditional adaptive logrank methods.