They can change the endpoints prior to unblinding.
They can change them whenever they want to. The issue is if an RA will consider the change to be a problem for a BLA/MAA.
The key rule is that the change cannot be based on information from the trial.
In this case the primary endpoint change was after the 2015 IA that saw the PFS endpoint was a failure (regardless of reason). As such, that change is dubious.
Further, they were looking for OS between the 233 and 99 as they announced they were waiting on the final 2 OS events. And then they went dark on that and said they were waiting on, well whatever, IMO, they clearly saw the failure of 233 vs 99 OS so again switched, this time to cross trial comps.
So, they have had 3 shots on goal. That is called multiplicity and serious issue.
Oh, and let us forget they knew both the blended curve and the data from the external controls prior to the analysis. Given there was no separation in the -L trial arms, kind of means they knew almost everything before they wrote the SAP.
Market Data 
Markets 
AI
