ABSTRACT
I. INTRODUCTION Estimation of the required sample size is a key issue for almost all clinical trials. Traditionally, there have been two distinct designs, fixed and sequential sample size designs (Meinert, 1986). Strictly speaking, a fixed sample size design is one in which the investigator specifies the required sample size before ~tarting the trial. A classical sequential design is one in which patient enrollment continues until the observed test-control treatment difference exceeds a predefined boundary value (Wald, 1947) or the number of patients exceeds a prespecified limit (Bross, 1952; Armitage, 1957). The drawback of a fixed sample design is that the sample size estimation depends on previous data or some guesswork of the parameters, which can be very unreliable. The use of classical sequential designs, on the other hand, is limited to situations where outcome assessment can be made shortly after patients are enrolled in the trial. For most trials in which long periods of follow-up are required, the so-called group sequential design has become popular [see, e.g., Lan and DeMets, (1983), O'Brien and Fleming (1979), Pocock (1977, 1982), Geller and Pocock (1987), and Gould and Pecore, (1982)]. However, the focus of group sequential methods has been on the early stopping rules of a clinical trial to fulfill
ethical concerns. These designs appear frequently in clinical trials involving cancer, myocardial infarction, and other type of life-threatening diseases, where mortality is the endpoint. For comparative trials, group sequential methods require that the treatment codes be broken at the interim stage and significance tests be repeatedly conducted on the accumulating data. Consequently, the type I error rate at each analysis stage has to be adjusted so that the overall type I error rate can be maintained at a predetermined level (Armitage et al., 1969).
