DICE: A Bayesian model for early dose finding in phase I trials with multiple treatment courses

Area of Science:

• Oncology
• Clinical Trials
• Biostatistics

Background:

• Phase I oncology trials traditionally define maximum tolerated dose (MTD) based on short-term toxicities.
• Existing model-based designs often overlook cumulative drug effects across multiple treatment cycles.

Purpose of the Study:

• To develop a Bayesian cumulative modeling approach to estimate dose-limiting toxicity (DLT) probabilities over treatment cycles.
• To propose the Dose-fInding CumulativE (DICE) design for identifying the maximum tolerated dose sequence (MTS).

Main Methods:

• Utilized a Bayesian cumulative modeling approach to account for cumulative drug effects and DLT occurrence cycle.
• Developed the DICE design for adaptive dose escalation and de-escalation based on observed toxicities.
• Conducted extensive simulations to compare DICE against the time-to-event continual reassessment method (TITE-CRM) and a benchmark.

Main Results:

• The DICE approach demonstrated superior or comparable performance in selecting the correct maximum tolerated dose sequence (MTS) compared to existing methods.
• Evaluated the predictive capabilities of the DICE design.

Conclusions:

• The DICE design offers an effective method for dose-finding in oncology, considering cumulative toxicities over multiple cycles.
• This approach enhances the precision of MTD sequence identification in clinical trials.