Lightrock what does that mean?
" It appeared to me that the way he was doing kaplan-meier projection was not stochastic in time. It seemed more like a monthly scoring that did not allow for difference in individual lifespan (enough)"
How do you come up with monthly scoring estimates if you only have rough averages for 1, 2, 3, 4 and 5 year overall survivals, considering that the yearly survival estimates are most likely more accurate than any monthly estimates within any of these years?
Given about 395 patients in each arm, with let's say a higher than average dropout (LTFU) of 40 patients (about 10%), we are still left with about 355 patients who after an average of 3 years on trial are estimated to be subject to a 45% event rate (55% overall survival). That 45% rate of attrition for 355 SOC patients would suggest that we would end up within the 3 years with already 160 events in the SOC arm (355X45%=159.75).
With 160 SOC eventing after 3 years on trial, and a weighted average of about 5 years on trial when the cumulative number of 298 events was reached, how many more events were added between the 3rd and 5th year on trial?
If only an additional 6% of the 355 post LTFU patients in the SOC arm had evented during the last two years preceding data lock, 21 more casualties would be added to the 160 who evented during the first 3 years and we would end up with 181 events in the SOC arm vs. 117 events in the Multikine arm.
That result is still far better than a 10% improvement for the Multikine arm.
