A must read rebuttal of the nonsense that has appeared these last few days of short desperation.
BTW the future predictions of survival from 2017 and 2018, recall if you make it to 36 months, then you have a median survival of xxx, allows you to plot future KM values with a strong degree of certainty. The 5 year data point is around 17-20% of the combined groups which only drops to about 15% at 7.5 years. Combined folks, so 25% survival for the treated at 5 years is quite likely. These trial people were not young so between 5-7.5 years could just be age related stuff.
Thanks to exwannabe for making me think so hard about KM predictability many moons ago. He posted much critique when I used the lazy approach which was wrong, but shut up when I posted the right methods.