Abstract/Summary

In the present healthcare climate, there is an urgent need to increase the efficiency with which novel therapies are evaluated. Multi-arm adaptive trials allow multiple treatments to be tested within a single protocol and offer the facility to respond to emerging data. Such trials allow treatment arms to be dropped or even added partway through the trial, directing resources to promising treatments. In this thesis, methodologies for two-stage adaptive trials with binary outcomes are explored, focussing on those approaches in which an intermediate outcome may be used for the purposes of treatment selection. Methodology for the multi-arm multi-stage approach developed by Royston et al. (2003, 2011), here denoted MAMS(R), is extended so that feasible and admissible trial designs may be obtained under the log odds ratio parameterisation. A simulation study suggests that these MAMS(R) designs perform favourably compared with the well-established combination method when a common outcome is monitored, but not when an intermediate outcome is incorporated. A proposal is made for increasing the efficiency and flexibility of MAMS(R) methodology by implementing conditional error calculations within a closed testing procedure. This approach allows the trial design to be updated at the interim analysis, resulting in gains in efficiency, particularly in trials where an intermediate outcome is used and where some promising treatments are dropped. The conditional error approach is then extended to offer the facility of adding a new treatment arm to an ongoing multi-arm adaptive trial. The procedure achieves good power, ensures Type I error rate control and performs particularly well if a new treatment arm is added when promising treatments have been dropped from the trial. Recommendations for using the new developments are given. It is hoped that this research will widen the use of MAMS(R) methodology in practice.