A very interesting snippet from the Philippines conference: at 32:07 mark Dr. Rahman says:
This is very important: some ad-hoc analysis I have done show that our trial is slightly underpowered.
Designing a trial is not an easy matter as one has to make some assumptions in regards to the expected results. A company with lots of money and lots of time in their hands designs the trial such that, at the end, there is no doubt the drug works (often meaning large number of total patients).
This has a cost, of course.
In a Pandemic it was necessary to design a trial that could demonstrate efficacy (and safety, of course) in a short period of time.
All of the above to understand why our trial had 390 patients and no more (or less). Now, the DSMB, looking at the 50% results determined that NO MORE were necessary. This is very good news and we all have understood that.
But, there is a difference between looking for the POWER compliance as opposed to looking for the efficacy. Of course, DSMB would have stopped it if this was not good in the sense that there was no hope to have a good level of significance (p-value), however, if our trial was underpowered to start with (only my opinion), the half-way results need to be very good for them to project power compliance at the end of the trial. This, in the context (outlined by DR. Rahman words) that when the trial was designed there was the opportunity to INCREASE the number of patients (i.e. adjust the power midway if necessary).
Let me be more specific: If we do ad-hoc analysis at 45 deaths, to achieve a power of 81% the p-value should had read 0.006482. Obviously the DSMB at that point stops and thinks: "if we continue with this tendency will we get 0.05 and 80% power at the end ??)"
Or, to give another example: let's assume the DSMB calculates a p-value of 0.017 mid-way (with implicit ad-hoc power of 70.1%) they might very well think as follows: "The drug will certainly finish with statistically significant results (p-value < 0.05) but the Power (probability of false negative) might be a bit low at the end. Let's increase the number of patients such that both are meet."
They didn't say anything.
You catch my drift ????. All I am saying is that if the 42 more deaths we had follow the same trend we will likely obtain an excellent statistical significance and a good power.
Quote:
"Had an interim analysis after 50% of the patients were enrolled to see if additional patients were needed to improve the power of the study. And was done after 50% of patients and the DSMB told us to continue without any modifications to the study."
This is very important: some ad-hoc analysis I have done show that our trial is slightly underpowered.
Designing a trial is not an easy matter as one has to make some assumptions in regards to the expected results. A company with lots of money and lots of time in their hands designs the trial such that, at the end, there is no doubt the drug works (often meaning large number of total patients).
This has a cost, of course.
In a Pandemic it was necessary to design a trial that could demonstrate efficacy (and safety, of course) in a short period of time.
All of the above to understand why our trial had 390 patients and no more (or less). Now, the DSMB, looking at the 50% results determined that NO MORE were necessary. This is very good news and we all have understood that.
But, there is a difference between looking for the POWER compliance as opposed to looking for the efficacy. Of course, DSMB would have stopped it if this was not good in the sense that there was no hope to have a good level of significance (p-value), however, if our trial was underpowered to start with (only my opinion), the half-way results need to be very good for them to project power compliance at the end of the trial. This, in the context (outlined by DR. Rahman words) that when the trial was designed there was the opportunity to INCREASE the number of patients (i.e. adjust the power midway if necessary).
Let me be more specific: If we do ad-hoc analysis at 45 deaths, to achieve a power of 81% the p-value should had read 0.006482. Obviously the DSMB at that point stops and thinks: "if we continue with this tendency will we get 0.05 and 80% power at the end ??)"
Or, to give another example: let's assume the DSMB calculates a p-value of 0.017 mid-way (with implicit ad-hoc power of 70.1%) they might very well think as follows: "The drug will certainly finish with statistically significant results (p-value < 0.05) but the Power (probability of false negative) might be a bit low at the end. Let's increase the number of patients such that both are meet."
They didn't say anything.
You catch my drift ????. All I am saying is that if the 42 more deaths we had follow the same trend we will likely obtain an excellent statistical significance and a good power.
26
