Multi-armed bandits backfill Bayesian optimal interval design.

Area of Science:

• Clinical Pharmacology
• Biostatistics
• Oncology

Background:

• Cancer Phase I trials aim to find maximum tolerated dose (MTD) and optimal effective dose (OED).
• Project Optimus guidelines emphasize dose optimization for subsequent trials.
• There's a growing need for efficient methods to add backfill cohorts for dose selection.

Purpose of the Study:

• To apply multi-armed bandit (MAB) algorithms for selecting dose levels for backfill cohorts in Phase I trials.
• To propose an MAB method integrating efficacy modeling for improved dose selection.
• To demonstrate and evaluate the performance of MAB algorithms in this context.

Main Methods:

• Utilizing multi-armed bandit algorithms for exploratory selection of dose levels.
• Developing a novel MAB approach that incorporates efficacy modeling.
• Conducting simulations to evaluate the performance of the proposed methods.

Main Results:

• Multi-armed bandit algorithms provide a straightforward and accessible method for dose selection.
• The proposed efficacy-informed MAB approach enhances the selection of optimal doses.
• Simulations demonstrate the effectiveness of MAB in guiding backfill cohort allocation.

Conclusions:

• Multi-armed bandit algorithms are effective tools for optimizing dose selection in cancer Phase I trials.
• Integrating efficacy modeling with MAB improves the identification of optimal effective doses.
• These methods support efficient clinical trial design and drug development.