Overview of model-assisted design for phase I dose-finding trials in oncology

Numerous model-based designs for dose-finding have been proposed to improve the phase I trials (e.g., increasing the accuracy to identify the maximum tolerated dose). However, algorithm-based designs (e.g., the 3＋3 design) are still dominantly used regardless of relatively poor performance due to the complexity of the implementation of the model-based designs. In this review, we introduce the model-assisted designs that have the simplicity of algorithm-based design and the similar performance to model-based designs, which are promising approaches to transcend the conventional framework. In addition, we comprehensively review the expansion of model-assisted designs to a complex area such as developments in dual-agent therapies; considerations of fast accrual issues or late-onset toxicities which would be observed in targeted therapies and immunotherapies; the exploration of optimal dose for efficacy and safety; and the other considerations (e.g., toxicity grades or historical data).