ABSTRACT
As indicated earlier, the adaptation or modification made to a clinical trial includes prospective adaptation (by design), concurrent or on-going adaptation (ad hoc), and retrospective adaptation (at the end of the trial and prior to database lock or unblinding). Different adaptation or modification could lead to different adaptive designs with different levels of complexity. In practice, it is suggested that by design prospective adaptation be considered at the planning stage of a clinical trial (Gallo et al., 2006), although it may not reflect real practice in the conduct of clinical trials. Li (2006) pointed out that the use of adaptive design methods (either by design adaptation or ad hoc adaptation) provides a second chance to re-design the trial after seeing data internally or externally at interim. However, it may introduce so-called operational biases such as selection bias, method of evaluations, early withdrawal, and modification of treatments. Consequently, the adaptation employed may inflate type I error rate. Uchida (2006) also indicated that these biases could be translated to information (assessment) biases, which may include (i) patient enrollment, (ii) differential dropouts in favor of one treatment, (iii) crossover of the other treatment, (iv) protocol deviation due to additional medications/treatments, and (v) differential assessment of the treatments. As a result, it is difficult to interpret the clinically meaningful effect size for the treatments under study (see also, Quinlan, Gallo, and Krams, 2006).
