In this talk, I will tell about an emerging topic of the clinical trial design – adaptive designs. We will introduce what the adaptive design is, and consider its one of the most common variants – multi-arm multi-stage (MAMS) design, in which several treatments are compared to the common control. We will discuss the construction of the design and the sample size calculation ensuring that type I and type II errors are controlled.

In many MAMS trials, however, the fixed equal randomization of patients is used. Motivated by ethical concerns, we will then consider an developing class of adaptive designs allowing for the skewed allocation of patients towards the most effective treatments – the response-adaptive designs. Specifically, we will consider the multi-arm bandits model as one of the response-adaptive designs. We introduce a newly developed response-adaptive design which is based on recent advanced in information-theory and compare its performance with the standard (non-adaptive) approach and the multi-arm bandit design.