Believe it or not, I do not find your core numbers for expected event that bad. Methodology is sound (unless I missed something). Some minor issues at the bottom of this post.
On major issues to bring up:
. The reporting of "we reached 154 (or whatever) events" is generally delayed by a month or two. This is because of the process.
. I suspect the placebo arm will come in with longer OS than your sample trial. Happens very often.
Quibbles:
. It would be a hell of a lot easier to read your key chart (that one with the yellow bar across the 12m group) if the 4th column was % survival as directly read off the prototype trial chart. I.E., 12 months = 39%
. I think you are sliding a month. I admit I always get sloppy/confused myself on this issue.
. I tend to see a few small errors that might be rounding. Would be easier to verify numbers if you addressed my first quibble. :-)
Regardless, thanks for clarifying your post. I still certainly disagree about what "33%" means though.
tradero, a few things to mix into that conversation:
1) yes PPHMs SUNRISE is best compared with Herbst et al. 2010. By the way that was MY BIG mistake in my simulations, I used the 5.4 months instead of Herbst 9.8 Months as the MOS and of course that resulted in MUCH TO EARLY result expectations.
2) The real centre opening curve is the one below. However the enrolment days must be considered because they allow us to estimate the # patients at any given days based on the fact that today we know how many enrolment days it took for the total 582 patients. The enrolment day curve vs centre opening curve would show the golf-stick versus patients enrolment potential.
3) Now that we know that it took 2Y and 1M to enrol all 582 patients, that is 92114 enrolment days for ALL 161 centres combined (not in table above), we can say that SUNRISE enrolled 0.0063 patients per enrolment day. Or 1 patients EVERY 158 enrolment days (about every 5 enrolment months). That is an approximation because some centres will have enrolled faster then others but the average will hold as we are not interested WHAT centre enrolled the patients. For us 'ANY' centre is good enough.
4) On DEC 12th, 2014 29262 enrolment days divided by 158 is 185 patients that should have been enrolled IMO and not the 101 in your table. That are about 92 Control Arm and 93 Bavi patients (or the other way around) because randomizing was 1:1.
5) Using Herbst et al. 2010 we are currently PAST the 9.8 months MOS for ALL of those 92 Control Arm patients, we are in month 13 since their start of treatment. And if Bavi works according set end-point expectations then a 2 month improvement would mean that the Bavi patients would expect to live 9.8+2=11.8 months. So we also reached the 'virtual MOS target' of PPHM for those 93 Bavi arm patients of 2014 by now since we are past month 13 if the 2014 patients would ALL have been enrolled in DEC.
6) All 2014 patients WERE NOT all enrolled in DEC, the were enrolled since JAN 2014 and are therefore all together between 12 and 25 months in the trial. Therefore assuming that all Ctrl Arm patients enrolled in 2014 evented is VERY PROBABLY correct. That would give us 92 evented patients in a MOS window of 9.8-3 [6.8] to 9.8+3 [12.8].
7) As explained here we need 173 events for the 1st look-in. And since 173-92=81 there would be 81 more events needed for the first look-in.
8) Where can those events come from? For starters from Control Arm patients enrolled since JAN 215. Those patients are currently between 1 and 13 months in the trial too. 582-173=409 patients of which 205 control and 204 Bavi arm patients. So 205/3=51 patients (those of JAN/FEB/MAR) could have evented because they are 10 months in the trial which is more then the 9.8 MOS. BUT the MOS is not an average hence it is not reasonable to assume they ALL 51 evented. Let's arbitrary take slightly more the 50%, say 30 events. Then 81-30=51 events still needed.
9) The last 51 events can only come from control arm patients dying as time progresses OR from Bavi arm patients enrolled since JAN 2014. So we are talking about 93 patients enrolled in 2014 and very few from those enrolled after JAN 2015 because they are only 12 months in the trial. If all Bavi patients enrolled in 2014 were evented TOO then we should have had the 1st look-in since some time because then we have currently 173+30=203 events and would only be 59 events away from the second look-in.
10) Now it becomes interesting. If the 51 needed Bavi arms patients would have evented then we ALSO would have had sufficient events for the 1st look-in and yet we know it didn't happen YET! So AT LEAST (93-51)+1=43 BAVI patients enrolled in 2014 MUST STILL BE ALIVE!!!! OR the Control Arm patients must have outlived the Herbst et al. 2010 statistics in a GRANDIOSE way (as happened with the control arm of the 1st ln NSCLC vs historical data).
11) #10 means that 43 patients, the day they day, will populate the MOS table on the RIGHT of the aimed for 11.8 months. And that is an extremely strong indicator that the end-points will be reached. The more that apparently the Bavi patients enrolled in begin 2015 have also NOT contributed to the eventing.
CONCLUSION ---------- PPHM knows the above and in MORE detail because they have one invaluable source that we do not have and which allows this puzzle to be laid-out. They have the EXACT enrolment dates for each patient (without ID double blinded trial) and the eventing totals (without patient ID). If you have that a lot of the guessing work disappears. That is why they are already busy with Avid III and seem not to hesitate to start other Doce+Bavi trials. They now they'll capture the SOC for NSCLC and Breast behind what IO can put down today on the same large foot-prints. AIMO.