Bayesian Phase I/II Biomarker-based Dose Finding for Precision Medicine with Molecularly Targeted Agents

Area of Science:

• Biostatistics
• Pharmacometrics
• Computational Biology

Background:

• Optimal dosing of molecularly targeted agents can vary based on individual patient characteristics, particularly biomarker status.
• Personalized medicine aims to tailor treatments to individual patients for improved outcomes.

Purpose of the Study:

• To propose a Bayesian phase I/II dose-finding design for personalized optimal dosing of molecularly targeted agents.
• To identify patient-specific optimal doses based on biomarker status and dose-by-biomarker interactions.

Main Methods:

• Employed canonical partial least squares (CPLS) to reduce dimensionality from dose, biomarkers, and their interactions.
• Modeled ordinal toxicity and efficacy using a latent-variable approach with extracted components as covariates.
• Utilized a utility function to quantify dose desirability and developed a two-stage dose-finding algorithm.

Main Results:

• The proposed Bayesian design effectively handles high-dimensional biomarker data and dose interactions.
• The latent-variable model successfully incorporates key features of molecularly targeted agents.
• Simulation studies demonstrated the design's good operating characteristics and high probability of finding the personalized optimal dose.

Conclusions:

• The developed Bayesian dose-finding design enables personalized optimal dosing for molecularly targeted agents.
• This approach effectively addresses the complexity of biomarker-driven treatment personalization.
• The method shows promise for improving treatment efficacy and safety in precision oncology.