I agree that 5/10 survival from the Montefiore coh
Post# of 155130
(Total Views: 688)
Posted On: 05/23/2020 6:12:46 PM
I agree that 5/10 survival from the Montefiore cohort is seems encouraging, but t-tests and binomial tests only make sense if you have good samples to work with. We really don't (yet).
I've been doing the same sorts of calculations in my head but have been also thinking about how reliable/unreliable they are given the garbage data that we have to work with. Yes, if we know a priori that p(mortality) = 0.88 for untreated patients like those in the Montefiore cohort, then I suppose your analysis makes some sense.
But it is completely unclear how the sample for that estimate of p_hat = 88% (not p) that you cite relates to the Montefiore cohort or the S Cal. cohort, and if it is even a little different, your p-values will swing wildly. Is the binomial p really 88%? Probably not. Maybe not even close. [https://www.npr.org/sections/health-shots/2020/05/15/856768020/new-evidence-suggests-covid-19-patients-on-ventilators-usually-survive --- docs estimating 20-50% mortality in Michigan]. The critical question is not so much what it is for some generic population, but what it is for the sample of patients in the experiment. We really have no idea.
I expect the p-values we see from actual, quality data will be somewhere between 0.001 and 0.4. I have seen some evidence to suggest it might be closer to 0.001 than 0.4, namely, the IL-6 levels drop pretty quickly after leronlimab is administered. A priori, this is a good sign because IL-6 is a primary cause of the killer inflammation. Also, clinical results with the IL-6 blocker, tocilizumab, have shown promise in France and China. Maybe leronlimab has additional benefits too, as suggested by Dr Patterson, but we still don't know.
In my opinion, it's crazy that a drug with a good safety profile and tantalizing anecdotal stories about effectiveness cannot be tried---except under extraordinary Big Brother oversight---even if the patient and his or her doctor wish to.
**************
P.S. I doubt very much that the FDA will be using a fixed and known binomial p in any analyses of the trial data!
I've been doing the same sorts of calculations in my head but have been also thinking about how reliable/unreliable they are given the garbage data that we have to work with. Yes, if we know a priori that p(mortality) = 0.88 for untreated patients like those in the Montefiore cohort, then I suppose your analysis makes some sense.
But it is completely unclear how the sample for that estimate of p_hat = 88% (not p) that you cite relates to the Montefiore cohort or the S Cal. cohort, and if it is even a little different, your p-values will swing wildly. Is the binomial p really 88%? Probably not. Maybe not even close. [https://www.npr.org/sections/health-shots/2020/05/15/856768020/new-evidence-suggests-covid-19-patients-on-ventilators-usually-survive --- docs estimating 20-50% mortality in Michigan]. The critical question is not so much what it is for some generic population, but what it is for the sample of patients in the experiment. We really have no idea.
I expect the p-values we see from actual, quality data will be somewhere between 0.001 and 0.4. I have seen some evidence to suggest it might be closer to 0.001 than 0.4, namely, the IL-6 levels drop pretty quickly after leronlimab is administered. A priori, this is a good sign because IL-6 is a primary cause of the killer inflammation. Also, clinical results with the IL-6 blocker, tocilizumab, have shown promise in France and China. Maybe leronlimab has additional benefits too, as suggested by Dr Patterson, but we still don't know.
In my opinion, it's crazy that a drug with a good safety profile and tantalizing anecdotal stories about effectiveness cannot be tried---except under extraordinary Big Brother oversight---even if the patient and his or her doctor wish to.
**************
P.S. I doubt very much that the FDA will be using a fixed and known binomial p in any analyses of the trial data!
(1)
(0)