Bayesian optimization for estimating the maximum tolerated dose in Phase I clinical trials.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Parametric models for maximum tolerated dose (MTD) estimation rely on assumptions about dose-toxicity relationships, which can lead to suboptimal dose selection if the model is misspecified.
• Existing methods often use point estimates, neglecting the uncertainty in dose-toxicity curves.

Purpose of the Study:

• To propose a flexible and robust Bayesian optimization method for estimating the maximum tolerated dose (MTD).
• To address the limitations of parametric models and point-estimate approaches in dose-finding studies.

Main Methods:

• The proposed method utilizes a Bayesian optimization framework with a nonparametric model for dose-toxicity relationships.
• It incorporates uncertainty quantification from the probabilistic model to guide dose selection.
• Dose escalation is managed with specific limitations to ensure safety.

Main Results:

• The Bayesian optimization method demonstrated successful estimation of the maximum tolerated dose.
• Simulation results showed accurate MTD selection and safe dose allocations.
• Performance was comparable to or better than existing Bayesian and nonparametric methods.

Conclusions:

• The proposed Bayesian optimization method provides a more flexible and reliable approach for estimating the maximum tolerated dose.
• This method enhances patient safety in clinical trials by accounting for dose-toxicity uncertainties.