Continual reassessment method with regularization in phase I clinical trials.

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacology

Background:

• Standard Phase I trial designs like the continual reassessment method (CRM) and modified toxicity probability interval (mTPI) have limitations.
• Long-memory designs (e.g., CRM) risk model misspecification, while short-memory designs (e.g., mTPI) may lose efficiency.

Purpose of the Study:

• To propose a novel dose-finding trial design, the regularized CRM (rCRM), that overcomes limitations of existing methods.
• To offer a design balancing the benefits of long-term and short-term memory approaches in early-phase clinical trials.

Main Methods:

• The regularized CRM (rCRM) is introduced, incorporating a regularization term to the standard CRM.
• rCRM assigns greater weight to data from doses with estimated toxicity probabilities closer to the target toxicity rate.
• This regularization shrinks model dimensions, enhancing the performance of the 2-parameter CRM.

Main Results:

• The rCRM design demonstrates improved performance compared to traditional CRM methods.
• Dose assignments in rCRM are less volatile than in standard CRM, particularly in expansion cohort phases.
• The method offers a robust trade-off between long-term and short-term memory design characteristics.

Conclusions:

• The regularized CRM (rCRM) is a promising design for Phase I dose-finding trials.
• Its balanced approach to memory enhances efficiency and stability in dose selection.
• rCRM is particularly suitable for guiding dose assignments in expansion cohort studies.