Types of Adaptive Design
Depending on the adaptations employed, Chow and Chang (2011) classified adaptive designs into 10 types: (1) an adaptive randomization design, (2) a group sequential design or an adaptive group sequential design, (3) a flexible sample size re-estimation (SSRE) design, (4) a drop-the-losers design (or pick-the-winner design), (5) an adaptive dose-finding design, (6) a biomarker-adaptive design (or enrichment design in a target clinical trial), (7) an adaptive treatment-switching design, (8) an adaptive-hypothesis design, (9) a two-stage seamless adaptive design (e.g., a two-stage phase I/II or II/III adaptive design), and (10) a multiple adaptive design. These designs are briefly described below. Detailed information regarding these adaptive designs can be found elsewhere (Chow and Chang 2008, 2011).
Adaptive Randomization Design
Adaptive randomization design allows modification of randomization schedules based on varied and/or unequal probabilities of treatment assignment, both prospectively and after review of the response of previously assigned subjects. The purpose is to assign more subjects to a promising test treatment under investigation and potentially to increase the probability of success of the trial. The commonly used adaptive randomization procedures include treatment-adaptive randomization, covariate-adaptive randomization, and response-adaptive randomization. In practice, adaptive randomization design may be valuable in trials with a relatively small sample size or trials with shorter treatment duration or short-term outcomes (e.g., biomarker or surrogate endpoint), but it may not be feasible for a large trial with a relatively long treatment duration.
Adaptive randomization is considered a less well-understood design according to the FDA draft guidance (FDA 2010a). In practice, it is often difficult, if not impossible, to characterize the probability structure of a clinical trial utilizing adaptive randomization design, especially when varied and/ or unequal probabilities of treatment assignment are used. In addition, the balance of patient characteristics between the treatment groups is a concern for this type of design. An imbalance in important characteristics is problematic, especially for confirmatory studies. In practice, an extreme imbalance may be caused by adaptive randomization at a relatively early stage of a trial. A typical example would be the Michigan Extracorporeal Membrane Oxygenation trial, in which only 1 out of 12 infants was enrolled to receive conventional therapy (Bartlett et al. 1985). As a result, Mugford et al. (2008) considered this trial to have a high potential risk for assignment bias.