Two-Stage Designs for Clinical Trials

Several papers recently have discussed the design of clinical trials from a gain function viewpoint (Anscombe [1963], Colton [1963, 1965], and, on a very similar topic, Wetherill and Campling [1966] among others). This paper considers the situation in which two treatments are available for a particular ailment. The response to the treatment is a continuous variable, and the gain in using a treatment is some monotonic function of the value of the response. A given number N of patients will be in need of one or other of the two treatments, and the trial is aimed at allocating the treatments to the patients in such a way as to make the expected gain as large as possible. The trial which achieves the largest expected gain is of considerable complexity, involving a decision with each patient as to which treatment to use, and usually one confines attention to trials with a simpler design. Colton [1963] considered single-stage designs and sequential probability ratio tests, and in a later paper [1965] two types of two-stage designs. In this paper, less restricted and unrestricted two-stage designs are constructed and evaluated. The model used here is the same as used by Colton, and is as follows: