I need new analysis conducted--- AVEO management please read.......
What I need to see from the statistician who is supporting this filing is to group patients into 5 Groups
Group 1 and 2 : those patients who took only one treatment ( Tivo or Sora) and did not take any second line therapy. This category will consist of 2 groups ( TIVO group and Sorafenib group)
Group 3 : Those patient who took Tivo and second line therapy
Group4 : Those who took Sorafenib and then Tivo as socond line therapy
Group5 : Those who tool Sorafenib and sewcond line therapy but not Tivo
Then conduct a survival analysis comapring these 5 groups and trying to
1. Comapre those who one one therpay only ,(TIVO versus Sorafenib)
2. Compare those in the Sorafenib group who took Tivo versus other 2nd line
How can we send this to the management at aveo? I think this will help enormously in showing the efficacy of Tivo as first line as well as second line.....
The OS subgroup analyses already reported are very close to what you suggest. They do not break out 2nd-line tivo vs. non-tivo after sorafenib, but most of them are getting 2nd-line tivo, so subdividing will not tell us much.
The real problem here is that it can be very misleading to subdivide a trial population based on things that happen _after_ randomization. Look at how Adam F. interpreted these OS subgroup results. The problem is that the groups are composed of very different patients, and are not comparable. Who are the patients getting 2nd-line in each group, are they sicker or less sick at the time of cross-over than those who do not get 2nd-line? Since we do not know the answer to those questions, it is very hard to interpret these subanalyses of OS in any useful way.
There are approaches that can address this issue. Pfizer, Novartis and others have used them to show positive OS results from trials that initially showed no OS effect for targeted therapies in oncology (GIST, pNET, RCC). If a message could be sent to management, it would be to look closely at what those sponsors did (the key methods are RPSFT and IPCW) and to have that in their presentation slides for the ODAC. Maybe they have already done it. If not, surely they scanning YMB for useful advice and can get these analyses done by Thursday AM. LOL
This is a good point you mentioned of using RPSFT (Rank Preserving Structural Failure Time Model) But also the analyses I suggested will shed some light on both goals at hand which is why I think mangement at AVEO took that route of crossing over patients who progressed on sorafening to Tivo ( First line and second line).
surely before establishing comaprability or even superiority based on overall survival, one needs to assess that patients are comparable among all these gorup based on demographics and basleine characteristics,,,,etc.
In one of their presentations, they slightly touched that and showed that there is some benefit for those who took only one treatment. I did not see that in their briefing document.
The RPSFT Model as you suggeted should also in theory adjust for the crossover issue and I am disappointed not to see this analysis presented, I am afriad that the statisitician working on supporting this submission may not be epxerienced enough,,,,,,
I didn't know you work for FDA. They know, more than you dude. Effectiveness is based on the number of patient successfully enrolled, not even 7 % enrolled. And FDA is not even sure these patients are from this country or cooked up results. Management has to answer the truthfulness here, otherwise go to litigations
If you go to the AVEO website and look at their OS poster (under investor relations), they have some of this information (if I am understanding you correctly). Unfortunately, they do have a chart comparing the subpopulation that had next-line VEGF therapy. That is, people who took Tivo then took next line VEGF (which Sora is one of) versus people that took Sora and then took another VEGF (mostly Tivo). That bad news is that the control arm (Sora) still had better OS. I assume the good news is that this data isn't statistically significant.