"I was wondering if a trial must be based on only ONE endpoint or could they have a trial that have both so that if the PFS did not meet their criteria, they could continue same trial to determine OS. It would allow having the chance of a faster approval but if the endpoint of PFS is not met, there would be no requirement to start another trial for OS."
Let's look at PFS and trial design for a second. In some indications PFS is not approval criteria, in some it is. For ovarian cancer, PFS is considered approval criteria as long as there is at least a trend toward improved survival. Sponsors typically like to be able to claim a survival advantage on the product label. Doing so requires at least a secondary endpoint for survival. When a trial has multiple ways to achieve an approvable result, the sponsor must account for the increased probability of a false positive by adjusting the statistical analysis to account for multiplicity. That is why F1 required p=.025 in the "high" subgroup in lieu of the more common p=.05 or less normally seen.
In Forward 1, in the high subgroup, MS was superior to control in both PFS and OS, but the OS improvement was by a wider margin. One can speculate that this may indicate a longer tail on the survival curve, ie the further out one looks, the better the comparative result. Knowing this, Forward 3 will definitely measure OS, but there are two ways to do it. If you want an approval based on PFS or OS, then you must make that multiplicity adjustment and the PFS analysis will require something much better than p=.05 or less. Or you can make PFS the sole primary endpoint, use the full p=.05 alpha availability and then test for OS further down the road for labeling purposes. The danger though is that if PFS misses, there is no do-over for OS.