When they amended the protocol to increase the size a bit and change the PFS trigger to 248, the also split the alpha between the primary PFS endpoint and the secondary OS enpoint. The study was now powered to detect a difference at 98% certainty (p=0.02) rather than the original 95% certainty (p=0.05). By doing so, they elevated the OS endpoint. There is your reason. Need I post the PR?
Please do link to the PR, because they have never disclosed they split the primary alpha.
You will come back with the PR describing the alpha of .02 for PFS (as it was prior to the resize), and then wave you hands and claim OS must be .03.