Point and interval estimation following a sequential clinical trial

The usual methodology employed in analysis after a sequential clinical trial is based on orderings of the possible samples resulting from the design. However, this approach lacks flexibility for use in wider applications. In this paper two estimation techniques not based on orderings are considered and modified to obtain improved accuracy. A bias-adjusted maximum likelihood estimate together with a new and general method for setting confidence limits are discussed. The realistic scenario of group sequential monitoring is assumed and methods for exact estimation are given. Accuracy of the methodology after a triangular test and an O'Brien & Fleming test are demonstrated through simulation.