Adaptive randomization methods for sequential multiple assignment randomized trials (smarts) via thompson sampling

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Adaptive Trial Design

Background:

• Response-adaptive randomization (RAR) offers ethical and statistical advantages in single-stage trials.
• Sequential Multiple Assignment Randomized Trials (SMARTs) are crucial for multi-stage treatment regime evaluation.
• RAR benefits remain underexplored within the complex SMART framework.

Purpose of the Study:

• To develop and evaluate novel RAR algorithms for SMARTs.
• To adapt Thompson Sampling (TS) for use in multi-stage adaptive trial designs.
• To ensure valid statistical inference for treatment regimes under RAR in SMARTs.

Main Methods:

• Proposed a suite of RAR algorithms for SMARTs, extending Thompson Sampling (TS).
• Developed post-study inferential procedures accounting for RAR's non-standard asymptotic behavior.
• Utilized empirical studies on real-world SMART data for validation.

Main Results:

• The proposed TS-based RAR algorithms improve in-trial subject outcomes.
• Efficiency for post-trial treatment regime comparisons is maintained.
• The algorithms are the first to address non-standard limiting behaviors in multi-stage adaptive trials.

Conclusions:

• Thompson Sampling-based RAR is effective for SMARTs.
• These methods enhance ethical trial conduct and patient outcomes.
• The developed procedures support robust statistical inference in complex adaptive trials.