Bayesian model averaging for partial ordering continual reassessment methods

Area of Science:

• Clinical Pharmacology
• Biostatistics
• Drug Development

Background:

• Phase I clinical trials are critical for advancing new therapies.
• Traditional dose-finding methods like the 3+3 and Continual Reassessment Method (CRM) struggle with uncertainty in combination drug trials.
• Combination trials involve simultaneous dose escalation across multiple drugs, creating complex dose-toxicity relationships.

Purpose of the Study:

• To address the 'estimation incoherency' issue in the CRM for partial ordering (POCRM) for combination drug trials.
• To introduce and formalize the Bayesian model averaged POCRM (BMA-POCRM) as a solution.
• To provide theoretical guarantees and demonstrate the effectiveness of BMA-POCRM.

Main Methods:

• The study analyzes the vulnerability of POCRM to estimation incoherency.
• Bayesian model averaging is employed to develop BMA-POCRM, considering all possible dose-toxicity orderings simultaneously.
• Theoretical guarantees for the estimation coherency of both POCRM and BMA-POCRM are derived.

Main Results:

• POCRM was found to be susceptible to illogical shifts in toxicity estimates, compromising patient safety.
• BMA-POCRM significantly reduces the frequency of estimation incoherencies by averaging over all possible orderings.
• Simulation studies and a specific case of POCRM incoherency demonstrate BMA-POCRM's improved performance.

Conclusions:

• BMA-POCRM offers a more robust and safer approach to dose-finding in Phase I combination drug trials.
• The method enhances accuracy and reliability compared to the standard POCRM.
• BMA-POCRM increases clinician trust in dose-finding models within complex trial settings.