Machine Learning Optimization February 2024

Bayesian Optimization for Clinical Trials

Applying Bayesian optimization strategies to rapidly and safely identify optimal dosing regimens in early-phase clinical trials.

Bayesian Optimization for Clinical Trials

Finding the optimal dose of a new drug that maximizes efficacy while minimizing toxicity is a classic challenge in Phase I/II clinical trials.

The Innovation

Instead of traditional rule-based escalation designs, we modeled the dose-toxicity and dose-efficacy surfaces simultaneously using Gaussian Processes.

The Algorithm

By employing Bayesian Optimization with custom acquisition functions designed to penalize highly toxic regions, our algorithm can conceptually identify the optimal biological dose with fewer simulated patients. This approach represents a significant step toward safer, more efficient adaptive trial designs.