Loved 😀how you guys struggled explaining an "almost statistically significant" finding. It is always tricky. Just remember that a confidence interval is not a probability. Here is how I would describe that finding. Would be happy to hear other approaches.
1) This is a single trial
2) The difference in stroke seen in this one trial was a 1.2% lower risk of the primary outcome in the early treatment group
3) If this trial had looked at the entire population then the true difference is 1.2% - but this trial is only a subset of the population - kind of like a poll
4) If one doesn’t look at the entire population then we use inferential statistics to help us “guesstimate” what the true result might be
5) The 95% confidence interval in this single trial was from -2.8% to +0.5%
6) All we can say is that if the trial was done 100 times 95% of the confidence intervals would contain the true difference - but we can never know what the actual difference is unless we study the entire population - a poll versus the actual election
7) When we only have one trial I think the best we can do is say the true difference is likely (and likely cannot be defined by a number or a probability) somewhere within the confidence interval reported in this one trial - but we could be wrong as this could be one of the confidence intervals that does not include the true result
8) So given that we only have one trial what should we do? Well assuming we rule out any undetected source of bias or confounding - then the best we can say is the result is likely somewhere between -2.8% and +0.5% - THEN the rest of the discussion has to focus around clinical judgment - the seriousness of the outcome, the size of the potential effect, the side effects, the cost etc - that is where all your important and valuable clinical skills come into play
9) If you know the confidence intervals, knowing the p-value literally adds nothing to the understanding of the results
Then there is Bayesian statistics - which is a whole other discussion.
Checkout my sensible medicine post on inferential stats for more details
If I may ask, will the last segment of Ask Us Anything round #2 not be published? Or perhaps never conducted? I suppose in either case owing to Vinay's new obligations? I am very keen to know if he has internalized my comment, as it couldn't be more relevant to his new job.
I'm not reassured about the early stoppages.
Loved 😀how you guys struggled explaining an "almost statistically significant" finding. It is always tricky. Just remember that a confidence interval is not a probability. Here is how I would describe that finding. Would be happy to hear other approaches.
1) This is a single trial
2) The difference in stroke seen in this one trial was a 1.2% lower risk of the primary outcome in the early treatment group
3) If this trial had looked at the entire population then the true difference is 1.2% - but this trial is only a subset of the population - kind of like a poll
4) If one doesn’t look at the entire population then we use inferential statistics to help us “guesstimate” what the true result might be
5) The 95% confidence interval in this single trial was from -2.8% to +0.5%
6) All we can say is that if the trial was done 100 times 95% of the confidence intervals would contain the true difference - but we can never know what the actual difference is unless we study the entire population - a poll versus the actual election
7) When we only have one trial I think the best we can do is say the true difference is likely (and likely cannot be defined by a number or a probability) somewhere within the confidence interval reported in this one trial - but we could be wrong as this could be one of the confidence intervals that does not include the true result
8) So given that we only have one trial what should we do? Well assuming we rule out any undetected source of bias or confounding - then the best we can say is the result is likely somewhere between -2.8% and +0.5% - THEN the rest of the discussion has to focus around clinical judgment - the seriousness of the outcome, the size of the potential effect, the side effects, the cost etc - that is where all your important and valuable clinical skills come into play
9) If you know the confidence intervals, knowing the p-value literally adds nothing to the understanding of the results
Then there is Bayesian statistics - which is a whole other discussion.
Checkout my sensible medicine post on inferential stats for more details
https://www.sensible-med.com/p/doing-statistics-can-be-difficult?utm_source=publication-search
If I may ask, will the last segment of Ask Us Anything round #2 not be published? Or perhaps never conducted? I suppose in either case owing to Vinay's new obligations? I am very keen to know if he has internalized my comment, as it couldn't be more relevant to his new job.
Dr. Prasad is out of the FDA.
Yikes. Probably chose his soul over the cesspool. Can't blame him.
Right? Laura Loomer had him in her cross hairs.