Rogue,
Thank you for you answer and again congrats for this fantastic piece of work
To answer in disorder to your questions.
Am I sure about those datapoints, the answer is that I am 90% confident as :
- it comes from an IRB minutes from which we had the links and original document
- I run my model with given efficacy to match Feb 3 2018 and when it did, it matched almost magically Feb 4 2017
- so far it gave me _ since January _ a predictive date for 298 (Q3/Q4) that has not been proven wrong so far while people were expecting early Q1 from Cel Sci or H2 from Ergomed
Also, to match those end point, I believe you would have to change your protocol a little bit, here is one method. It might be a bit lengthy though, so up to you if you want to spend time on this !
I don't know if you would like it as it has to assume some efficacy (ps I'd rather use Taïwan soc as in my static simulation it matches pretty well) :
1- consider two KM Life tables, one with SOC (Taïwan) and the other with soc + x % and 5% dropouts.
2- Setdate to endpoint date (for intance 3feb 2018 first)
3- run the stuff 10 000 x to compute most probable number of total (in the two arms) events in that date
4- adjust x so that most-probable number of events falls around 208
5- perform a reality check on 4 feb 2017 : run the stuff and calculate most probable number of events. Adjust x a little bit if required
6- Run the stuff and see which dates 298 is falling in 10 000 runs
7- Publish to us probability of having date between for instance D-15 days D+15 days, D being most probable date
8 - This should give us x expected efficacy and D date with a % of certainty
9 - Re-run at will with different dropouts rates etc... have some fun !
Regarding your last sentence, yes you are right, but considering that we don't know the reality, there are still three parameters that can lengthen the end date : soc OS, efficacy, dropouts. I'd rather prefer efficacy than the 2 others. On soc I believe we are close to reality. On Dropouts, its complete darkness except that we have an upper limit.