Update on planning of phase II clinical trials.

A review is given of the statistical aspects of plans for phase II cancer clinical trials. The following plans were reviewed: a minimum number of patients plan proposed by Gehan in 1961, such that if no responses are observed, further study of the agent can be discontinued; a two-stage decision theory approach proposed by Sylvester and Staquet in 1977; a limited patient accrual plan proposed by Lee et al. in 1979, which assumes the number of patients available to be fixed, a false negative error to be more serious than a false positive error, and the possibility that the ultimate result will be inconclusive; a predictive probability plan of Herson, which permits early stopping when the initial results are poor; and a one-sample multiple testing procedure of Fleming, which assumes the trial to be carried out in k stages (usually k = 1, 2 or 3), with a test performed at each stage that permits early stopping when results are either very favourable or unfavourable. When the true response rate (p) is either very low (p less than p0) or in an area of definite interest (p greater than pa), the plans proposed by Gehan, Lee and Fleming perform reasonably well and an investigator should choose the plan that best fits the circumstances. None of these plans work very well when the response rate is intermediate (p0 less than p less than pa). The plans of Sylvester and Staquet require much more prior information, including data on expected utilities, but their plans can be optimal when such information is available.(ABSTRACT TRUNCATED AT 250 WORDS)