...We Welcome You To The Resverlogix HUB withIn The AGORACOM COMMUNITY!

Free
Message: Trials and statistics, next steps

J-bot

Very interesting article, albeit a little bit of a heavy read but still great find nonetheless!

In reading the article, it seems like what they are saying is that for P Values that are higher or less certain, you actually need quite a bit of bolt on data to move the dial.  In other words it's not linear from what I gather.   

And this is because...if the premise of the initial trial is that you need lots of patients and hours to draw a proper conclusion in the first place, and if that conclusion results a high P Value - the higher the greater uncertainty or risk of chance, and so if you want to do a bolt on trial to make it more reliable, the only way you can essentially add reliability (for lack of a better term) is to do so by adding lots of additonal patients and hours. Else your bolt on trial and data set is inherently unreliable to begin with.   

In other words, by bolting on only a small fractional amount of patients, that small additional bolt trial will have on its own, a very high amount of inherent variation due to its small size to begin with, which in essence creates further drag on the original trial data set when combined.

Thus, for high P Values, it seems to me that you basically have to repeat the trial with similar sized numbers as the original trial and hope for a different result which could be the definition of insanity. :).

So if I am at all in the ball park here which I very well may not be, for RVX, their P Value just missed the mark so maybe they could actually get away with slightly smaller bolt on numbers.  

Or perhaps the best course of action of this is not the case and if they are required to do so, would be to completely redo a new P3 trial as it might take the same amount of statistical powering as a bolt on trial anyway, and in doing so, they should remove stroke and add whatever other good results/factors into the new P3 baseline.  And if this is so, a good partnership or buyout would be the way to go.

Then again i could completely be wrong here and if so please cut me some slack. This was a masters level article!

10BagR

Share
New Message
Please login to post a reply