(Total Views: 798)
Posted On: 07/02/2020 7:41:59 PM
Post# of 153903
Re: Coocooburra #40716
As a statistician, I have been concerned the S/C placebo death rate might be below 40%, depending on the S vs. C mix. This issue substantially impacts statistical power. NP's comment on "many deaths" reassures me your 40% guess is plausible-to-low.
With 120 patients split 80/40 - yesterday's enrollment - in the interim review and deaths of 40% in placebo and 20% in treatment, then p=.0127. Increasing sample size to 150 - say two more weeks - yields p=.006 (one sided, difference in proportions test - neither may be correct).
As you know, a lower placebo death rate makes it harder to prove, say, a 50% reduction in rate in the drug arm, as below @ 120 patients:
Placebo deaths 50%, treatment 25% has p=.003
Placebo deaths 30%, treatment 15% has p=.035 (if 2-sided test, not significant)
With 120 patients split 80/40 - yesterday's enrollment - in the interim review and deaths of 40% in placebo and 20% in treatment, then p=.0127. Increasing sample size to 150 - say two more weeks - yields p=.006 (one sided, difference in proportions test - neither may be correct).
As you know, a lower placebo death rate makes it harder to prove, say, a 50% reduction in rate in the drug arm, as below @ 120 patients:
Placebo deaths 50%, treatment 25% has p=.003
Placebo deaths 30%, treatment 15% has p=.035 (if 2-sided test, not significant)
(2)
(0)
