Required duration and power determinations for historically controlled studies of survival times.

This paper provides power formulae for historically controlled studies with survival time endpoints, and considers two designs. The first design involves no further follow-up of control group subjects; the second design involves collection of additional information on control group subjects who had not failed at the time of design of the prospective portion of the study. The power functions differ from those for concurrently controlled studies, since, under the first design, the control group data are constant for purposes of power calculations, while under the second design, some of the control group data are constant while the remainder are random. I contrast the required duration of these studies with the required duration for comparable concurrently controlled studies under a wide range of conditions.