Comparison Among Modified Continual Reassessment Methods with Different Dose Allocation Methods for Phase I Clinical Trials

The continual reassessment method (CRM) has been an essential Bayesian finding design in phase I clinical trials. It utilizes all the information in observed data which contributes to its essential operational characteristics. However, the CRM has been criticized for its aggressive dose escalation. Model-assisted methods including BOIN, Keyboard, and mTPI improved the safety while retaining relative efficiency. In this paper, we propose four models combining the structure of the CRM and model-assisted methods. We show that these models could operate with comparable CRM performance through simulations. The results suggest that two of the proposed methods outperformed the traditional methods with a higher percentage of correct selection of true maximum tolerated dose. In addition, the interval-based approaches offered by the new models with greater flexibility regarding target toxicity achieved an improvement in the adaptability of the dose-finding process in clinical trials.