Adaptive designs for best treatment identification with top-two Thompson sampling and acceleration

Area of Science:

• Clinical Trials
• Biostatistics
• Machine Learning

Background:

• Adaptive clinical trials are crucial for efficient drug development.
• Identifying the best treatment arm early maximizes patient benefit.
• Existing multi-armed bandit (MAB) approaches focus on identification error rates.

Purpose of the Study:

• To adapt MAB approaches for clinical trial settings, prioritizing cumulative patient benefit.
• To introduce and evaluate Top-Two Thompson Sampling (TTTS) and a novel variant (TTTS2) for small sample sizes typical in drug development.
• To enhance the performance of adaptive designs in identifying superior treatments.

Main Methods:

• Utilized Top-Two Thompson Sampling (TTTS) and a simplified variant (TTTS2).
• Developed an acceleration approach for TTTS tailored to smaller sample sizes.
• Conducted extensive simulations to assess performance in typical drug development scenarios.

Main Results:

• TTTS and TTTS2 demonstrated effectiveness in identifying the best treatment arm.
• The proposed methods showed strong performance in small sample settings.
• The acceleration approach improved performance in drug development contexts.

Conclusions:

• TTTS and TTTS2 are effective and practical methods for adaptive clinical trials.
• These approaches enhance the identification of superior treatments while increasing cumulative patient benefit.
• The proposed methods offer advantages for early-phase drug development with limited sample sizes.