A number of designs of phase I dose-finding trials have been developed. Algorithm-based designs such as standard 3 + 3 designs are easy to understand and implement since they do not require explicit model specification for a dose-toxicity relationship. On the other hand, model-based designs such as the continual reassessment method (CRM) (O’Quigley et al., 1990) have been proposed. The author will give a review of the CRM and its related topics. In particular, the author makes mention of some of the problems with 3 + 3 designs that have often been used in phase I dose-finding studies and gives a detailed description of ideas, concepts, theories, properties and issues in the CRM.
During the last two decades there have been many methodological developments in the design and analysis of phase I, phase II and phase I/II dose-finding studies. Much of these developments originate from the continual reassessment method (CRM), which has gained popularity since its first proposal by O’Quigley et al. (1990). In this paper the author will give a review of the CRM and its related topics. In particular, the author takes a general view of dose-finding and -escalation designs that are extended from or related to the CRM, providing a brief bibliographical introduction of them.
Recently, it is becoming more challenging to cope with low success probabilities, much cost and severe competitions in a new drug development. An adaptive design is considered to be a promising tool for efficient development for its various types of adaptations and applications in multiple development stages. However, its flexibility does not necessarily make the development more efficient and it may always allow a risk of operational bias to use such a design. Thus, to consider the specific benefit and risk of each adaptive design using quantitative measures under a concrete setting, it is important to make intensive discussions and to share experiences between each side persons in charge. In this review, focusing on sample size re-estimation which is a relatively simple and basic adaptive design, I will give some outlines of methods and review statistical points to consider.