Improved adaptive randomization strategies for a seamless Phase I/II dose-finding design

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacology

Background:

• Seamless Phase I/II designs are crucial for efficient dose-finding in cancer drug development.
• Existing adaptive randomization (AR) methods may not optimally balance dose-efficacy assumptions, especially for molecularly targeted agents.
• The Wages and Tait (2015) design, while innovative, has limitations in model selection and sample size specification for AR.

Purpose of the Study:

• To propose and evaluate novel randomization strategies for the AR stage in seamless Phase I/II dose-finding designs.
• To enhance the estimation of optimal doses and improve patient allocation to effective treatments.
• To address limitations of the original Wages and Tait design, including arbitrary sample size requirements.

Main Methods:

• Developed three alternative randomization strategies calculating probabilities based on candidate model likelihood.
• Compared proposed methods against the original design using extensive simulations.
• Evaluated performance based on optimal dose estimation accuracy and patient allocation to effective doses.

Main Results:

• The proposed methods, particularly the recommended strategy, demonstrated superior performance in simulations.
• The recommended method improved patient allocation to the optimal dose across most scenarios.
• Enhanced accuracy in selecting the final optimal dose was observed without a significant increase in toxicity.

Conclusions:

• The proposed randomization strategies offer a significant improvement over existing AR methods in seamless Phase I/II dose-finding.
• The recommended strategy provides a more robust and efficient approach to dose escalation and de-escalation.
• These findings have implications for optimizing cancer clinical trial designs and accelerating drug development.