Event-based analysis times for randomised clinical trials

Analysis times for clinical trials are often based on the time when a speci fied total number of events occur. A modification is proposed in which this time is based on when either arm reaches a prespecified number of events. The implications in terms of the expected number of events required to achi eve the same size and power at a specified alternative are explored. Saving s are achieved when alternatives for which there is reasonable power hold, but more events are required at the null hypothesis.