I hope I can help on the power front. Some ideas e
Post# of 153917
(Total Views: 645)
Posted On: 11/06/2020 9:42:21 PM
I hope I can help on the power front. Some ideas expressed have been inaccurate. Maybe we can learn a bit about CD-12 from the Humanigen situation.
Power, precisely, is the probability of a false negative - the probability that there is an underlying effect in the larger population, but you didn't see it because there were too few enrollees to have the signal dominate the noise (i.e. p>.05).
Power calculations is thus depend on (1) how big an effect you want to detect, and (2) how confident you want to be of signal dominating noise.
Repeat: you can't calculate power until you specify the alternative. You can calculate required sample size from a fixed alternative and desired power, or calculate power from a fixed alternative and known sample size.
My best read on the Humanigen reporting is that the target hazard ratio is 1.29, the desired power is 90% and that, given the data already received (I think 201 recoveries), they were told, rather than the planned 56 more recoveries, they needed nearly 200 more recoveries to reach that.
It's a not-too-difficult matter to back-calculate the number of events that must have already happened in the treatment group, and that's where they got their current 1.37 ratio.
Can we apply this to CD-12? NP knows the target difference we proposed and the power specified in calculating the 390 patients - we don't. We know that the required N to maintain that power at the interim was no greater than 390.
So it seems that Amarex could do a similar back-calculation and come up with a minimum difference in mortality in the groups at the interim. So maybe that's why no one was willing to pay a penalty to unblind the trial.
Power, precisely, is the probability of a false negative - the probability that there is an underlying effect in the larger population, but you didn't see it because there were too few enrollees to have the signal dominate the noise (i.e. p>.05).
Power calculations is thus depend on (1) how big an effect you want to detect, and (2) how confident you want to be of signal dominating noise.
Repeat: you can't calculate power until you specify the alternative. You can calculate required sample size from a fixed alternative and desired power, or calculate power from a fixed alternative and known sample size.
My best read on the Humanigen reporting is that the target hazard ratio is 1.29, the desired power is 90% and that, given the data already received (I think 201 recoveries), they were told, rather than the planned 56 more recoveries, they needed nearly 200 more recoveries to reach that.
It's a not-too-difficult matter to back-calculate the number of events that must have already happened in the treatment group, and that's where they got their current 1.37 ratio.
Can we apply this to CD-12? NP knows the target difference we proposed and the power specified in calculating the 390 patients - we don't. We know that the required N to maintain that power at the interim was no greater than 390.
So it seems that Amarex could do a similar back-calculation and come up with a minimum difference in mortality in the groups at the interim. So maybe that's why no one was willing to pay a penalty to unblind the trial.
(10)
(0)