Phase I trial design for drug combinations with Bayesian model averaging.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacology

Background:

• Selecting statistical models for two-dimensional dose finding in drug-combination trials presents a significant challenge.
• Existing models often exhibit varying design properties, with no single model universally outperforming others.

Purpose of the Study:

• To comprehensively compare four distinct dose-finding methods in two-dimensional drug-combination trials.
• To introduce and evaluate Bayesian model averaging (BMA) as a robust approach to address model selection uncertainty.

Main Methods:

• A comparative analysis of four dose-finding methods using the same underlying algorithm across different model structures.
• Extensive simulation studies were conducted to assess operating characteristics in diverse practical scenarios.
• Implementation of Bayesian model averaging by assigning prior probabilities to each model and adaptively allocating patients based on posterior toxicity estimates.

Main Results:

• Simulation results indicated that different statistical models yield distinct design properties.
• No single model demonstrated superior performance across all tested scenarios, highlighting the need for a more flexible approach.
• The Bayesian model averaging approach proved robust across various simulated scenarios, outperforming individual models.

Conclusions:

• The choice of statistical model significantly impacts the properties of two-dimensional dose-finding designs in drug-combination studies.
• Bayesian model averaging provides a robust and adaptive strategy to mitigate the arbitrariness of model selection.
• This approach enhances the reliability of dose-finding designs, particularly in complex clinical trial settings.