Optimizing Dosing Strategies in Cell Therapy With Machine Learning and Exposure-Response Integration.

Area of Science:

• Biotechnology
• Clinical Trials
• Pharmacometrics

Background:

• Cell therapy efficacy and safety are linked to cell expansion, which varies due to patient and product heterogeneity.
• Integrating exposure assessments is crucial for informed dose selection in cell therapy development.
• FDA's Project OPTIMUS emphasizes comprehensive data for optimal dose determination.

Purpose of the Study:

• To propose a seamless Phase I/II clinical trial design for optimal cell therapy dose selection.
• To integrate toxicity, efficacy, cellular kinetics (CK), and patient/product characteristics into a data-driven dose-finding process.
• To leverage random forest (RF) modeling for guiding dose escalation and identifying the optimal dose (OD).

Main Methods:

• A seamless Phase I/II design integrating toxicity, efficacy, CK, and baseline characteristics.
• Utilizing random forest (RF) modeling with comprehensive data for dose escalation and narrowing options.
• Implementing interim analyses based on RF estimations to discontinue futile or toxic doses.
• Randomly assigning additional patients to promising doses for further evaluation.

Main Results:

• The proposed RF-based design accurately selects the optimal dose (OD) with desirable operating characteristics.
• The design effectively allocates patients to potentially therapeutic doses while minimizing exposure to toxic doses.
• Simulation studies demonstrate high accuracy in OD selection, with the algorithm selecting 0-3 doses on average for exploration.

Conclusions:

• The RF-based seamless Phase I/II design offers a data-driven, efficient method for optimal cell therapy dose determination.
• Integrating exposure data and patient/product characteristics enhances the precision of dose selection.
• This approach aligns with regulatory expectations for optimizing cell therapy development and patient safety.