Here is something different for you guys.
First, the assumptions
Assume the shape of the kaplan meier PFS curves for the control arm (RFA only) in the HEAT trial will look like the one in http://www.ncbi.nlm.nih.gov/pubmed/21445671, which is nearly the same as the ones in other HCC/RFA research papers.
Also, assume that the PFS curve of thermodox+RFA will look like the PFS curve of TACE+RFA. This is a reasonable assumption as the mode of operations of both are similar.
Notice that I'm talking about the shapes of the curves and not absolute values. Whether the HR of thermodox+RFA is as good as TACE+RFA is what the trial and hopefully before that a monte carlo simulator will tell you using the date of 190th event.
That's pretty much it for the assumptions.
>>this thread is so long and unruly, would you please just post the one (are there more?) post that explains those figures?
Again, she used 15 months in the "proof", but we can now [probably] use 16 months:
>>BTW, here is how you can verify that 15m pfs should lead to early unblinding.
1) Assume that the 190th event occurs within a week (by 6/21/2011)
2) 50% of the patients were enrolled before 3/21/2010 (you can check the enrollment data).
3) There are 15 months between 3/21/2010 and 6/21/2011.
4) Assume all the patients were enrolled on 3/21/2010 (this overestimates the performance of thermodox arm; that's why we need to use a monte carlo simulator).
5. If the median PFS for the control arm is 15m and since the control patients were enrolled for a median of 15months, 50% of the control patients enrolled would have progressed by 6/21/2011 (this is the definition of median)
6. There are 300 control patients
7. 50% of 300 control patients is 150 control patients
8. If 150 of the events of 190 event is caused by the control patients, the rest, 40, is caused by the thermodox patients.
9. This leads to an HR of 40/150 = 0.27, which is way better than the 0.75HR required by the SPA.
However, neither the enrollment nor the PFS curves are straight, that's why I use a monte carlo simulator and arrive at a more correct (and more conservative) estimate of ~0.5HR which may still lead to early unblinding if the control arm has a PFS of 15months.
Again, this is math. How much probability you assign to co's 12m estimate or not is your call.
Whoa!! I thought I read this thread. Where does it say the following ???:
" we can basically rewrite that now with 7/15 being 16 months for the proof -- meaning 150 placebo events and 40 TDox events!"
150 placebo and 40 Tdox is... well... that's f'in statistically significant! Trond, this thread is so long and unruly, would you please just post the one (are there more?) post that explains those figures? Thank you!
This fell way too far down the list. Re-reading this thread gave me goosebumps....
Summer, consistantly great analysis.
Westwell, you are a demi-god -- late July was indeed 600 enrollment and I will promote you to full deity if July 15th was the 190th. In one of the subposts, summer gives a "proof" based on enrollment at 6/11 being 15 months.... we can basically rewrite that now with 7/15 being 16 months for the proof -- meaning 150 placebo events and 40 TDox events!
:-P Maybe I need to calm down. lol...