I thought we needed something other than Ebola.
So food for thought...
The IMPRESS trial control would "expect" median survival of around 20 months.
Our middle patient for the overall trial (both groups) was enrolled June of 2012, so including 8 weeks for surgery has been "in trial" for 28 months. We know the company is projecting 333 events by the end of this year, so since Feb at 222 events, we are clocking at about 10-11 events per month. This would also mean that the 361st event would be projected for a March time frame. (about 7 months from now) putting the overall median survival for this trial up around 35 months. (28 current time in plus 7 more)
If control is down at 20, then that would put trial at more like 50 months, wow.
ok, back to my corner
I for one would appreciate your on-going posts. I've hired a consultant to look at the publicly available trial information. He specializes in drug trial results with a further specialty in oncology. His perspective is this looks very promising, which is about all I want to say at this point.
There is always a strong tendency to extrapolate a desired result given one or two data points, but the tendency should be strongly resisted especially when your hard earned money is involved. What you are doing is akin to taking a scatter diagram and choosing only two points, projecting a straight line and then concluding you can predict the final result. This ignores the fact that there are almost endless possibilities that can be projected from numerous combinations of all the other points.
What is needed to get a reasonable estimate of the test group are: 15-20 dates and numbers of enrollment patient data, 10-15 points of event data, an understanding that OS data has a non-linear, right-skewed distribution, chi-squared comes to mind. With that data we could perform Monte Carlo simulations using 15, 18, 20, 22,...,50 median month OS rate distributions until we could match the enrollment data with the event outcomes.
Then we would have OS survival for the total patient population. Then we could then assume the control group had a 20 median month OS rate and run simulations again until we determined an OS rate for the test group that matched enrollment and event data.
We could also run simulations with increasing control group OS rates to see how good the control group would have to be to negate the statistical significance of the trial.
If you have the data and a command of math and statistics this is not that difficult. However, without the data we are in a fog. Limited data combined with Hokus-pokus math plus smoke and mirror logic will not yield satisfactory estimates. Do not risk your money based on a hopelessly flawed mathematical model. Cheers and good luck to you.
PaulAnth: This is such a good reasoned response on the statistical modeling aspects of the pIII trial, and I want to thank you. It is substantive, thought provoking, and even handed. We need more of this type discussion on this board, and less chatter about how the stock is going to perform, or the latest on ebola - in statistical parlance, just plain background noise. You could apply another principal to the mean OS of the survival group, which is improved techniques and treatments over time, which moves the MOS past 20 months. But, barring a large jump in progress, the MOS for the control if it moves is more likely to move from 20 to 21 or 22, than it is from 20 to 28 for example. This isn't a definitive, but given the history of slowly creeping MOS for pancreatic cancer, surgery, and therapy, it is a good likely scenario. Given the data we can only guess at for now, it appears in any case the MOS of the treatment in pIII looks to be a positive. I'm mostly heartened by the fact that Genentech made a $150mm up front payment, after looking at the non-publicly disclosed data.
Paul Anthony, It could be as simple as counting on fingers or as complicated as developing a bells and whistles model, but I think its worthwhile using the data we have to assess the likelihood of success at the 2nd interim. Afterall, this is what the big holders, with a LOT of hard earned money at stake are doing. Its good to follow your nose, if you have one.
All that being said I agree with some of your sentiments. Playing around with numbers can help you to reinforce a certain narrative you are telling yourself about how well you think the trial maybe going. Others go in the opposite direction and use generalist statements - like "well no cancer vaccine has really worked well until now - so why should this one?". The truth speaks eventually. Just a few more months to wait.
BTW. For those who havent noticed already NLNK have a positive trait of surprising with results from work that has been kept under wraps for some time. All I am saying here is with all the IMPRESS/Ebola chatter dont forget the PILLAR, SSCLC and IDO BC trial - all due for topline results at end of year. Not far away.
Please read through old posts on this board. Most of what you're suggesting has been done by several people. Several enrollment milestones are published or known and have been used to extrapolate enrollment. I believe everyone who has posted projections has used historical K-M curves for control and have based trial estimates on published phase 2 results.
I have decided to stop making projections myself for the very reasons you suggest, but that does not mean that, taken with a grain of salt, the general trend can't be approximated or appreciated based on the consensus of projections.
It sounds like you may have some background. Please feel free to do the work and put your estimates out there rather than just criticize others who have put the work in.