howdy, lorbas. Maybe it's just fortuitous happenst
Post# of 155071
(Total Views: 724)
Posted On: 06/07/2020 2:31:11 PM
howdy, lorbas. Maybe it's just fortuitous happenstance, but in the s/c trial, if the number of fatalities with leronlimab is less than those with the placebo, the p-value is less than 0.05. If the mortality rate for placebo is high (>60%, but depends on sample size), then the number of fatalities under leronlimab can exceed the number under placebo and still have p < 0.05.
For a reasonable instacheck:
if leronlimab has the same number or fewer fatalities than the placebo, p-value <= 0.05 in all cases (this means that the empirical fatality rate for leronlimab is half of what it was for placebo...could be what NP meant by his comment that each placebo fatality counts double or something to that effect).
If the mortality rate for placebo is high, it's a little more complicated, but here's what leronlimab needs to do as a function of the number of placebo fatalities (given a total of n = 75 patients with 2:1 ratio leronlimab to placebo) to get p-value <= 0.05 (two-sample, one-sided test on logits)
placebo, leronlimab:
16, <=18
17, <=20
18, <=22
19, <=25
20, <=28
The m/m trial is not as simple because it is on an ordinal scale from 0 to 12, and I have no idea what the variances will be or what distributions would make sense for the test statistics or modeling the data (quite possible they will do a non-parametric test, which makes speculation about results even more difficult).
For a reasonable instacheck:
if leronlimab has the same number or fewer fatalities than the placebo, p-value <= 0.05 in all cases (this means that the empirical fatality rate for leronlimab is half of what it was for placebo...could be what NP meant by his comment that each placebo fatality counts double or something to that effect).
If the mortality rate for placebo is high, it's a little more complicated, but here's what leronlimab needs to do as a function of the number of placebo fatalities (given a total of n = 75 patients with 2:1 ratio leronlimab to placebo) to get p-value <= 0.05 (two-sample, one-sided test on logits)
placebo, leronlimab:
16, <=18
17, <=20
18, <=22
19, <=25
20, <=28
The m/m trial is not as simple because it is on an ordinal scale from 0 to 12, and I have no idea what the variances will be or what distributions would make sense for the test statistics or modeling the data (quite possible they will do a non-parametric test, which makes speculation about results even more difficult).
(3)
(0)