Extension of Fisher's least significant difference method to multi-armed group-sequential response-adaptive designs

Area of Science:

• Clinical Trials
• Biostatistics
• Experimental Design

Background:

• Multi-armed multi-stage designs are used in clinical trials for evaluating experimental treatments against a control.
• Response-adaptive randomization enhances these designs by enabling early stopping and prioritizing promising treatments.
• Existing frequentist designs can be conservative and lack power, especially for distinct therapies, and use fixed allocation ratios.

Purpose of the Study:

• To investigate Fisher's least significant difference method extended to group-sequential response-adaptive designs.
• To demonstrate that information accrual continues after dropping inferior arms, allowing for error-spending approaches to control family-wise error rate.
• To compare the operating characteristics of the proposed adaptive design with fixed-sample designs for normal and survival outcomes.

Main Methods:

• Mathematical derivation showing information accrual continuation after arm dropping.
• Application of error-spending approach for family-wise error rate control.
• Simulation studies and redesign of the NeoSphere trial to evaluate group-sequential response-adaptive designs for normal and censored survival outcomes.
• Consideration of two optimal allocation strategies: one for efficient estimation and one for maximizing power.

Main Results:

• The proposed group-sequential response-adaptive design controls the family-wise error rate effectively.
• The adaptive design demonstrates efficient and ethical advantages compared to fixed-sample designs.
• Simulation results confirm the favorable operating characteristics for both normal and survival data.

Conclusions:

• Group-sequential response-adaptive designs offer significant advantages in clinical trials.
• These adaptive designs provide a more powerful and efficient alternative to traditional fixed-sample designs.
• The family-wise error rate is well-controlled, ensuring statistical validity in adaptive trial settings.