Havasu, You are absolutely right. Numbers do m
Post# of 153880
(Total Views: 762)
Posted On: 09/14/2020 8:17:06 AM
Havasu,
You are absolutely right. Numbers do matter. The larger the better. However, as you can understand, the larger the more time and money it takes.
Maybe not an issue for GILD but definitely an issue for us or any other pre-revenue company.
So we need to probe that LL works with the minimal amount of people to demonstrate two things:
1) The probability that we think the drug works but we think it doesn’t (power of the trial). Normally this should be 80% or 90. This is called “false negative” error or type II error.
2) The probability that LL doesn’t work (the null hypothesis). This is reflected in the p-value and is normally taken as 0.025 for two tailed tests. In another words, there is a 2.5% percent probability that good results are due to chance (the drug does not work but we think it works).
So, as you say, the more patients, the higher the power and the higher the possibility that, it the trial says it works, it works.
This is more problematic when the trial is using randomization schemes with unequal allocation ratios such as the one 2:1 (LL vs placebo) we are using as this requires 12% more patients than a trial using 1:1 to detect the same size effect with equivalent power.
The FDA is obviously well aware of all of this and gave us a number that could fulfill their comfort level of power and statistical significance in case we can produce results.
Now, let’s assume we just miss the p-value of 0.05 (or 0.025) … is that the end ??? imo unlikely, we are in a pandemic and there is secondary endpoint that can tell the FDA: hey !!! this drug is working.
Bear in mind that in an interim analysis the number of patients is one half of the trial design (that is less power), so, we need to produce stellar results to demonstrate right there that LL works. Otherwise we will need to finalize with full enrollment.
But I think we will take it out of the ball park.
BTW: The study mentions 27% SURVIVAL not mortality.
Read More: https://investorshangout.com/post/view?id=589...z6Y1C5AnpM
You are absolutely right. Numbers do matter. The larger the better. However, as you can understand, the larger the more time and money it takes.
Maybe not an issue for GILD but definitely an issue for us or any other pre-revenue company.
So we need to probe that LL works with the minimal amount of people to demonstrate two things:
1) The probability that we think the drug works but we think it doesn’t (power of the trial). Normally this should be 80% or 90. This is called “false negative” error or type II error.
2) The probability that LL doesn’t work (the null hypothesis). This is reflected in the p-value and is normally taken as 0.025 for two tailed tests. In another words, there is a 2.5% percent probability that good results are due to chance (the drug does not work but we think it works).
So, as you say, the more patients, the higher the power and the higher the possibility that, it the trial says it works, it works.
This is more problematic when the trial is using randomization schemes with unequal allocation ratios such as the one 2:1 (LL vs placebo) we are using as this requires 12% more patients than a trial using 1:1 to detect the same size effect with equivalent power.
The FDA is obviously well aware of all of this and gave us a number that could fulfill their comfort level of power and statistical significance in case we can produce results.
Now, let’s assume we just miss the p-value of 0.05 (or 0.025) … is that the end ??? imo unlikely, we are in a pandemic and there is secondary endpoint that can tell the FDA: hey !!! this drug is working.
Bear in mind that in an interim analysis the number of patients is one half of the trial design (that is less power), so, we need to produce stellar results to demonstrate right there that LL works. Otherwise we will need to finalize with full enrollment.
But I think we will take it out of the ball park.
BTW: The study mentions 27% SURVIVAL not mortality.
Quote:
27% mortality in the placebo arm would be around 16 or 17.
Read More: https://investorshangout.com/post/view?id=589...z6Y1C5AnpM
(5)
(0)