Overwhelming efficacy - the stats required, page-282

I have been reading up on this and I think what we are used to thinking about interim read outs (using frequentist methodology) is quite different to how it should be considered compared to an interim read out using Bayesian theory ...

the p value < 0.01 or <0.001 method is very binary... you either hit that or you don’t... and interim read outs are are actually only done once at most twice in a trial... and usually much later in the trial when enough patients have recruited to make the p value achievable.

Using Bayesian methodology- you can do many interim read outs and much earlier in the trial.. which is the reason why this is used- primarily it can save a sponsor a lot of time and money.

As long as you have set up your trial correctly and your product works.. Bayesian theory will use the patient data at hand, looking at many variables and modelling out how the rest of the trial patients will fair based on the data at hand.

So in my mind- it basically extrapolates the result of you trial based on data from your first 90 patients... and it is more likely to predict an outcome that would result in an early stop for efficacy, using a lot fewer patients than if you used your p value / frequentist method.

In short- if remestemcel-l really works... this first interim analysis has a much better shot than what I first thought it would... because using a frequentist method of thinking, doing an interim analysis at only 30% of recruitment would be almost impossible for an early stop. Not so- for Bayesian theory.