Response-adaptive trial designs with accelerated Thompson sampling

Area of Science:

• Clinical Trials
• Biostatistics
• Machine Learning

Background:

• Optimal patient allocation to treatments is crucial in clinical trials, particularly for rare diseases.
• The Gittins index rule offers optimal sequential allocation but is complex and lacks randomness.
• Thompson sampling balances allocation and randomness but shows poor short-run performance in clinical settings.

Purpose of the Study:

• To enhance the short-run performance of Thompson sampling in clinical trial patient allocation.
• To introduce a novel, easily implementable acceleration approach for Thompson sampling.
• To adapt the method for batch allocation scenarios and improve patient satisfaction in preference trials.

Main Methods:

• Development of a novel acceleration approach for Thompson sampling.
• Evaluation of the approach through simulation studies.
• Application to the redesign of a patient preference trial.

Main Results:

• The proposed acceleration approach significantly improves Thompson sampling's performance.
• Enhanced average total response rates were observed in simulations.
• The method is applicable to batch allocation and simplifies implementation.

Conclusions:

• The novel acceleration approach effectively improves Thompson sampling for clinical trial patient allocation.
• This method offers a practical solution for optimizing treatment assignment and patient satisfaction.
• The approach is versatile, applicable to both sequential and batch allocation designs.