Phase II studies in oncology are frequently conducted as two-stage single-arm trials with a binary endpoint indicating tumor response. As a common feature of these designs, the sample sizes of the two stages and the decision rules for the interim and the final analysis have to be pre-specified and adhered to strictly during the course of the trial in order to assure control of the type I error rate. In practice, however, the attained sample sizes often deviate from the planned ones leading to the situation of overrunning or underrunning. The currently available approaches to deal with this problem are either based on assumptions that are rarely met in practice or do not guarantee that the significance level is kept. However, strict control of the type I error rate plays an important role also for single-arm cancer trials, as they are frequently a fundamental part of the registration information. We propose a general methodology that allows handling both unintentional and intentional overrunning and underrunning while strictly controlling the type I error rate. Application of the proposed procedure and some of its characteristics are illustrated with a real phase II oncology trial.
Keywords: clinical trials; conditional error function; discrete data; overrunning; phase II; two-stage design; underrunning.
Copyright © 2015 John Wiley & Sons, Ltd.