Overwhelming efficacy - the stats required, page-179

Great thread.Thanks for getting it going.

I'm not sure whether this has been posted previously but the most useful information I could find on Bayesian analysis for early stopping was this article:

The utility of Bayesian predictive probabilities for interim monitoring of clinical trials

The authors say this in relation early interim analyses for efficacy:



I have assumed that the detailed protocol (which is almost certainly confidential) will include pre-determined p-values & other metrics governing stopping for overwhelming efficacy and that the Bayesian analysis will be used to estimate the likely outcome of the trial in the event of an early stop based on the data of patients who have already survived/died at the 30 day point as at at the data cut off date and (here's where the Bayesian analysis comes into play) to estimate the outcome for those who have already enrolled as at the data cut-off date but have not yet reached the 30 day mark. Any decision to unblind the data and terminate the trial is then made on the basis of that combined analysis, but with the final results taking into account the data of all patients enrolled at the cut-off date. So with recruitment now running at full throttle, I would expect this approach to add considerably to the body of information that informs the decision to either continue or stop early.

If anyone thinks I am mistaken here then I am very happy to be corrected,