Below is my rough projection of events by time; comments are welcomed.
Assumptions:
1. accrual is complete by end of June 2007, resulting in a total of 47 months of accrual.
2. 1% annnual loss to followup- resulting censored patients
3. median OS based on pooled 9901 and 9902a are 23.2 vs 18.9 months for provenge vs control, respectively. Since 9902a appears to have a worse prognostic than 9901, the pooled median may be representative of a mixture of patients due to change in protocol.
4. exponential distribution
Event times:
End of Deaths
06/2007 200
12/2007 250
06/2008 300
12/2008 330
06/2009 360
12/2009 383
06/2010 402
Discussion:
1. If the event target is 380 (at an alpha of 0.05), then an interim efficacy analysis of OS shortly after FDA approval, can be based on as much of 52% information fraction (200/385), the nominal alpha is 0.004, per lan demets method with an OBF boundary.
2. However, if using the pooled hazard ratio of 1.5 from 9901 and 9902a, to power the study at 90% would require only 256 events, which means the final analysis can be done by end of Dec 2007, and data release in 2008.
3. A conservative powering approach would be to use HR observed in 9902a, ie 1.3, but that would require 456 deaths to have 80% power, therefor this senario is not likely.
