Monitoring a three-armed clinical trial with survival endpoints: Fisher's least significant difference approach.

Researchers have long recognized the importance of monitoring trials to determine whether to terminate a trial early or change a trial because of a substantial treatment effect. Fisher's least significant difference (LSD) procedure has been suggested by Proschan et al. [Proschan, M. A., Follmann, D. A., Geller, N. L. (1994). Monitoring multi-armed trials. Stat. Med. 13:1441-1452] to control the overall type I error rate for trials with three or more arms and survival endpoints. In this paper we propose an alternative Fisher's LSD interim monitoring procedure that uses the same adjusted significance levels for the pairwise tests as for the global test; it continues performing a global hypothesis test at looks subsequent to a look where a global null hypothesis was rejected. We also examine revised Proschan et al. approach that uses the O'Brien-Fleming two-armed trial boundaries for pairwise tests. A simulation study shows that compared to the approach proposed by Proschan et al., the procedure we propose is more powerful and produces overall type I error rates closer to the nominal values when all groups are truly equivalent. For the scenarios examined, the overall type I error rates and power of the proposed approach are virtually equivalent to those of the revised Proschan et al. approach.