Monthly enrollment data for this trial is readily available via IMUC's Investor tab on their webpage. It is one of the slides in the October 2012 Presentation. I was able to construct a model of the trial based on that enrollment and the quoted historical survival data for standard of care. (61% 1yr, 26.5% 2yr, 15% 3yr). Let me just say that a ZERO benefit should have tagged the interim analysis point somewhere last July or August and would be FINISHING the trial about now. In my opinion, this looks pretty good even after I inflate the control group numbers.
I'm encouraged.
another way to look at the data regardless of where you project it to be is that every 10 days delay in the trigger adds about 1% to the survival time.
Great work by everyone here modeling possible outcomes. However I would like to point out a big caveat. Many are assuming a median survival of 14.6 months in the control arm. This figure was derived from the lankmark paper by Stupp et al. in 2005. I would like to caution everyone as median survival times have increased since then. I would therefore encourage people to use a 16-17 month median survival time as these are figures achieved in more recent publications. Additionally median survival significantly depends on the median age of the population, tumor size, extent of surgical resection etc which are all huge prognostic factors. We have all seen bad outcomes (e.g. CLSN) when using models assuming outcomes consistent with previous cohorts. Would just like to point this out as someone in the medical community who performs GBM research and has 17+ publications. Nonetheless, I am fully confident that we will see great data from IMUC, NWBO, and CLDX in the coming months/years.
Sentiment: Strong Buy
Agree with the caution in using models. In CLSN we're not sure yet if the models were way off due to assuming outcomes consistent with previous cohorts. On the conference call they said the control was about 20% better than expected, which should not have impacted the models to the extent the predicted results were from the actual results. There had to be other issues too, so it's even MORE dangerous in doing models than just potentially erroring on the control estimate.
Thanks for providing the UCLA info. Although I havent checked on this, if the mean age of the IMUC trial is 54 years, then can perhaps use the UCLA info for those aged (50-70; mean = 60). The median survival there was 556 days (~18 months). So theoretically IMUC's control arm can exceed that. Additionally, from what I remember, control patients are receiving SOC followed by investigator's choice at progression. Keep in mind that especially since IMUC's trial involves a lot of tertiary academic institutions, patients are going to be treated aggressively. This includes the use of multiple chemotherapeutic regimens including avastin, etoposide, irinotecan etc. Some patients experience robust effects. Additionally even in characterizing the response to SOC, patients vary depending on MGMT promotor methylation status. Also depending on how many tumors in the treatment and control arms achieve gross total or near total resection, survival can vary significantly. This will depend on tumor location, proximity to eloquent structures, multifocality etc. If there is significant tumor burden following resection, the vaccine may not be as effective as GBM tumors can have immunosuppressive effects. Hence CLDX's trial was limited to those who had minimal tumor burden. Im not saying IMUC's vaccine won't have a robust effect - I think all of the evidence points to it being efficacious. However using models assuming 14-15 months may be wishful thinking. I would be interested if someone can model 32 events using control survivals of 14, 15, 16, 17, 18, and 19 months.
Sentiment: Strong Buy
Good point. There is a big assumption about the placebo arm when running these numbers. That is why they randomize these trials.
If placebo OS is closer to 17 than 15, then you want to push back this timeline roughly two months. In that case, by now there would probably at least be a weak signal of efficacy but I would feel really good making it through this month and next month.
UCLA has a website with graphs for survivability based on age. I think one can just Google "UCLA GBM". The ave. age of IMUC's trial is 54. The UCLA figures were compiled in 2006, or thereabouts.
Of all the survivability figures I've found, 14.6 mos is the longest I've seen. Since the treatment for GBM hasn't changed appreciably, I'm not sure that no. has gotten much better. The only possibility for improvement would be with better resection techniques.
Sentiment: Buy
jet, bdoglo, jlwbig,
great discussion, thanks. could you guys share in a little more detail how you came up with your math as some people say zero efficacy kicked in in November ?
also when I look up the enrollment data on page 19 of the Oct presentation I just see those bars on the chart but no real accurate figures by month. did you look at the same chart or have other source ?
thanks for some more input on your stat models.
I just looked at the chart and I took the enrollment numbers from past press releases to get enrollment numbers. I then took about 44% of the enrollment numbers as being on trial, assigning 1/3 of a person to the placebo vaccine, and 2/3 to the active vaccine. That gets you your trial population.
The next part is a little more tricky, you need to estimate the placebo survival curve. You can use the one in the presentations from the Temodor trial in 2005. You then can reduce the number of people who started the placebo in each month for subsequent months according to the survival curve. The difference between the number of patients who started the trial and how many are left is your expected number of events for the placebo arm.
Then do the same thing on the active vaccine population. You can use the same survival curve to see what would happen if the active vaccine was the same as the placebo. I take the placebo curve values and raise them to some power less than 1 to make the curve extend out over time. This allows me to experiment with various curves by only modifying a single parameter. You can find a value that gives you a curve that is pretty similar to the phase I results for example.
Another alternative to simplify the survival curve is to assume there is a constant attrition rate where every month the current population is reduced by a constant rate (i.e. 5% of the population is lost each month in the placebo arm). This simplifies the model and you only have to deal with one parameter (the 5% for example). But this curve overestimates events early and underestimates later, probably because the surgery causes a measurable delay in progression and there seems like a small sub-population that creates a long tail on the survival curve (perhaps they are fully resected with a good margin).
The various approaches produce different numbers but they all seem to be within a few months of each other.
I also did a similar analysis except I fit survival to a constant attrition rate model so I can interpolate and extrapolate it more easily. Zero benefit would have been 32 events by about August (as chicken said) and 64 by this coming April (in my version). If the results came out tomorrow then I would estimate we will see about a 43 month median survival (Assuming it is about 15 months in the control group). So it is looking like Phase I results will be more strongly validated with this larger population size.
I concur having done my own study. I figured if we made it thru 2012 with no top line we are golden..as is the patient population..hope those with this disease see some good..GLTA
Sentiment: Buy