Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
<The recent abstract from Dr Preylak shows that in the subgroup that received Taxotere after provenge treatment survival was almost doubled over monotherapy. Did all the increase in survival benefit in the provenge arm come from this subgroup.>
Recall that the CD54 data were measured before Provenge was reinfused to patients. The high upreg patients clearly outperformed the low ones. Since chemos & Taxotere treatments were independent decisions without any knowledge of the CD54 data, it would be extremely unlikely that the high upreg group somehow coincided with the Provenge+Taxotere group.
All that does is to say that there were 30 patients on the 9902a. So the 2:1 distribution factor would say that 20 (instead of 19) on the treatment arm in 9902a took Taxtotere. Now, we'll be adding 30+20 instead of 31+19 to come out with the same 50 on the treatment arm in the integrated data who took Tax. So, no, the estimates will be unchanged. Of course, Dendreon has the exact numbers so they can give us the actual results.
Estimates of medians, 36-months survival rates, etc.
I just posted this on ivillage:
http://www1.investorvillage.com/smbd.asp?mb=971&mn=13214&pt=msg&mid=810619
Thanks PGS!
I wasn't able to listen in to the CC this morning. This is really helpful.
<<<The "25 months" number was more likely a construct of frustration than fact.>>
Of course it was not a construct of fact and more of a guestimate!>
That's a bad estimate given that with near certainty, patients will die in 48 months and the survival distribution is heavy tailed. Say if you are looking at a simple exponential distribution, then half would die at median, half of the remainder at two medians, etc. If you assume median survival is 25 months, there must be many still alive at 48 months. Btw, 16 months is about the placebo medians in 9902a and the Tax trials.
"I am not sure you can draw that conclusion because for a time 9902b was strictly enrolling less sick patients with Gleason scores <=7. Therefore, I am not sure if the final patient make-up would actually be sicker than the 9901/9902a population or actually less sick, in which case your 16 months may actually be 25 months or more..."
The Tax trials which included a wide range of patients show that the maximum survival time for HRPC was likely less than 48 months. Since these survival distribution curves tend to have long tails, it is unlikely that the median survival time would be more than 24 months. The "25 months" number was more likely a construct of frustration than fact.
Not sure how the censored data help to figure out the enrollment rate. Could you explain?
A reason to believe a trial is back-loaded is that usually not all trial centers are opened when the enrollment starts. So as more centers come online, more patients can be signed. This should have been the case with 9901 but should not be the case with 9902b - at least not since it was altered last year.
<The company wouldn't know which Provenge preparation was for a placebo arm participant or a Provenge arm participant. They would therefore measure CD54 in all prepared doses and ask for a new leukapheris from any failed batches.>
The expectation was not to see CD54 upregulation for placebos. However, it is likely that they would prepare a small sample to see if the collected blood did the right thing before freezing it for later manufacturing for the cross-over treatment. This would make sense since they are running a parallel phase-2 open trial to test the safety and effectiveness of the frozen blood. So yes, no information leak on asking patients to come back.
http://clinicaltrials.gov/ct/show/NCT00170066?order=2
The enrollment pattern for 9902b would not be as back-loaded as in other trials. A large number were already enrolled when they switched to all GS and all 70 trial centers were up and running at that time. So from late 2005 to whenever enrollment ends in 2007, the enrollment rate should be steady with maybe some spikes around times with heavy publicity. With sicker patients, the median OS will not be as robust as in 9901, assuming some benefit for cross-over, probably around 16 months for placebos. So assuming a 75% event goal, getting results in 2008 isn't a stretch.
after they messed up a reported
p-value it may be worth doing.
When did they mess up reporting some p-value and what on?
If they could show
the cd54 upreg in placebo patients and show
that the survival benefit only survives when you
actually reinfuse the upregged cells, then I
would be proven wrong but I haven't seen that yet.
Unless I am missing something, the above statement is a bit weird. Placebos were reinfused with their own blood untouched per trial protocol. But if you change "placebo" to "treated" in your statement, didn't the announced top-line result say what you want? Ie, the higher the upreg rate, the better the survival chance.
<I have finally found the motivation for the etchy-sketchy exercise of measuring the KM curve. Have you done this already?>
No. But if you end up doing it, please share the data. Thx.
CD54 upregulation is just one of the many metrics that they use to measure the potency of the product and the product can be made for anyone having PC or not. So it isn't clear how it could be used as surrogate endpoint. But wouldn't the statement on analyzing it as a continuous variable yielding a significant correlation with survival more or less satisfy what you are asking?
<Would managed care/CMS pay for an expensive procedure that was tantamount to a diagnostic for more than 75% of the patients?>
The data that we have seen only talks about CD54 upregulation as a potency measure of the product at the time that it is manufactured. Let's not build on PGS' hypothesis that CD54 upregulation is a persistent patient characteristic that you can use for diagnosis yet until we see real data to support it.
<But I have to say that it tends to back up my claim that Provenge is a rather weak immunostimulant given that only 38 patients met this criteria. Imagine what the overall survival survival data would look like if 76 met the criteria, for example?>
Note that the PR yesterday mentioned that the new analysis treated CD54 upregulation as a continuous variable. The p-value given was .011 which was likely much larger than that from the Provost chart comparing the populations above and below median. The important part is that the new analysis says that there is a clear correlation in the level of upregulation to survival.
At what upregulation level that potency is achieved is still not yet published - at least not where we can see it. But if you read the transcript of the FDA workshop where the early data were presented, Dr. Provost mentioned that there were cases when the product did not meet their threshold so they asked the patients to come in again and the results were fine second time. This says that CD54 upregulation is not always a persistent characteristic of the patient like a gene or some such that you can use to say if a particular patient would or would not benefit from the drug. It is mostly just a potency measure of the product at the time it is made.
E.g. Ocyan's hypothesis that Dendreon knows when people came in for active P-11 boosters. Or the company might have other blood markers for those booster patients vs the 'boosters' for placebo patients (e.g. PSA Velocity). We have access to 10 pieces of data - Dendreon has access to 100's of pieces of data (none conclusive, but still a lot more info than we have)
In the same vein of knowing about the #'s of placebo and active boosters, it is very likely that they already have the ellapse times of the median and 75% percentile for the placebo population. The difference between these gives a reasonable estimate of the median PSAR. If that matches with historical median PSAR in various studies using Lupron and similar chemical castration drugs, i.e., between 6 to 8 months, that would be good to moderately good news. On the other hand, if that is much longer than 8 months as some have conjectured due to the single Lupron shot, that would be not so good news - yet.
I am curious as to why you seem to think that it is a negative if certain analyses for survival were done in D9901 during looks at the TTP and pain endpoints. The protocol said to follow for 3 years for survival and there was no provision to stop early on that. So it shouldn't matter how many of those they did as they never got multiple hits for survival.
They might have tried for AA if TTP was met and that might have been thought of as another hit on the data and a cause to reduce the alpha for the survival endpoint. But even that consideration is questionable as current thinking is that alpha need not be penalized for dependent endpoints.
My results are empirical, not theoretical. Therefore I cannot say with absolute certainty that they are universal.
As long as the covariates considered are balanced, this is likely true in general and due to the noise filtering effect inherent in the iterative regression steps. There is quite a bit of discussion in various papers about how postieri corrections can increase the power of a study. See, for example, the section "When are adjustments made?" in the below:
http://www.tufts.edu/~gdallal/adjust.htm
In a survival analysis, median is defined as the point where the curve crosses the 50% point on the probability axis. Without censoring, that coincides with the normal median. But with censoring, it can shift to the right of the normal median by a significant amount. So the 10 people censored before the previous median of 25.9 months would shift the curve right quite a bit resulting in the current larger median.
This is actually the same issue with Dr. Gold's previous statement about the survival rates at 24 months of patients in the Taxotere trials. He forgot that significant censoring was going on in the Tax trials so looking only at the number of survivors at a certain time relative to the starting number would be wrong. It is good that he has dropped any reference to that in recent presentations.
<So if we rank the 69 remaining uncensored patients in the Provenge arm, Patient #35 (50th percentile) lived 35.2 months after randomization. We know that there were 28 patients who lived the full 36 months, meaning that there were 6 more who lived >=35.2 months and <=35.9 months.>
Wall - note that the non-PC-deaths were censored not excluded. So your estimate of 6 more who lived >=35 months would be too high. In fact, you can read that off the 9901 survival curves in ASCO 2005. Just eyeballing says that a more accurate estimate would be about three.
New data highlights.
1. The data got much better after censoring non-PC-related deaths. The p-value improves to p=.002 instead of p=.01. This p-value means that the chance that Provenge is a fluke is 1 in 500! And this is straight log-rank, no Cox regression involved.
2. Halabi nomograms show that there were no imbalances in prognostic factors. This uses historical data so it's complementary to the use of Cox regression as previously done with the 9901 data itself.
3. Chart 18 absolutely destroys the early chemo argument for treated patients. That chart shows the times when patients started chemotherapy. The Provenge curve is clearly above the placebo curve. That reinforces the notion that people naturally avoid chemo as long as they feel that their health remains good. People treated with Provenge certainly behaved that way.
4. Chart 26 shows what the label will look like for Provenge. It will be frontline for AIPC before any chemos. A point was made that this is not to make chemo obsolete. That likely means that Dr. Petrylak's presentation next month will further show that chemos after Provenge substantially increase the probability of survival.
http://investor.dendreon.com/downloads/22568dndn.pdf
Anyone pointing to the Avastin/FDA example in CRC as a likely parallel of Provenge's BLA path without also pointing to the Avastin/FDA example in adjuvant breast cancer use is being misleading. Furthermore, IMO, that misleading is deliberate.
Wall - care to expand on the deliberation part? Are there some nefarious motives beyond the usual over-enthused cheerleading?
6. I still believe that Dr. Raj Puri of CBER, who has been involved with DNDN since the beginning, will probably be the lead FDA doc. The CTGT panel’s Chair, Dr Mule’, an oncologist and immunologist, who, I believe, worked at the FDA at the time of DNDN’s initial Ph 1 trial, will remain and may have a substantial say in the appointment of any temporary members.
These are safe assumptions. Puri is the the Director of the Tumor Vaccine Division so it would be mighty strange if he is not directly involved in the first vaccine to come up for approval. I was told that the CTGTC was recently renewed for another two year term. Mule' is still listed as its Chair.
These are the characteristics of the Avastin trial that may give it some advantage over Provenge:
1. Avastin+Chemo over Chemo, not drug vs. placebo.
2. Large trial, 900 patients with an out-of-the-park p value.
3. Likely doing well in various subgroup analyses.
The main problem is 3 with the Provenge trials because 9902a is a subgroup of the integrated data that fails the log-rank test. We all know about the arguments on a smaller and abruptly truncated trial, imbalances, etc. However, it remains a controversial data point. A mitigation factor is that the Cox p-values are significant in all three data sets 9901, 9902a and 9901+02a. Further, the Cox p-value of 9901+02a is significantly smaller than of the others lending credence to the assertion that 9902a supports 9901. So an enlightened statistician may look at this data favorably.
But to me, the evidence pointing toward a panel is that its possibility must have been raised in the pre-BLA meeting and communication should have been going on with the company. As the company continues to expect a panel, that likely means that they have more info about that than publicly available. The below link is the published timeline to prepare for a panel for both the FDA and the company. Note that the need to convene a trial should be identified by the FDA 9 months beforehand and communicated to the sponsor. So if a panel is expected in Feb'07, that need should have been identified since last May/June. The company may hope that one will be waived but they should be working hard on preparing for it now.
http://www.fda.gov/cder/Offices/OTC/Abbreviated_timeline_FDA_CHPA_seminar_fall_06.pdf
Somebody posted this on iVillage:
SIPULEUCEL-T IN ADVANCED PROSTATE CANCER
Daniel Petrylak, M.D.
http://www.sinaionsitehealth.org/tcf/pdfs/symposium_agenda.pdf#search=%22petrylak%20sipuleucel-t%22
Dr. Petrylak will be talking on Nov. 10, likely about the use of Taxotere after Provenge.
patients with stable disease at the time of unblinding were allowed to cross over.
This is an unusual design - at least to me since the statisticians I work with like uniformity in designing their experiments. How did you find out about this?
That’s true but irrelevant in this case because the PFS endpoint was hit with so much p-value to spare.
What you know post facto does not alter what you should plan a priori. They could have just taken out an insurance. After all an earlier phase-3 on Satraplatin was capped early by the original developer (BMS?) when an interim look was not impressive.
What GPCB did was to increase the number of PFS events in the unblinding trigger beyond the value originally specified in the SPA.
Then, this would have no effect on the survival result. Patients still cross over whenever they progress. What this did do was to increase the power for the PFS endpoint itself since there were more events to work with (less censored patients).
Re: Satraplatin (addendum)
GPCB’s Bernd Seizinger handled this trial smartly by maneuvering to delay the unblinding beyond the point where it was originally scheduled to occur. By so doing, he delayed crossovers from the control arm to the Satra arm and thereby improved the probability that Satra will eventually show a decent (although not necessarily statsig) benefit in overall survival.
I hope you are not saying that he delayed giving Satraplatin to placebo patients after progression to improve the chance of getting survival significance. This would be unethical to say the least and possibly illegal.
As the trial was event based and blinded, it would take some interesting intervention to have any influence on the time for unblinding. On the other hand, if the drug worked better than expected for the sample of patients, then a delay beyond projection would be natural. The DNDN's P11 trial may be going through this experience.
Rancherho - There is no question of the feeling of being lucky/blessed when such a clear-cut result as the CD54 upregulation data was seen. The important part in all this is that they have gone through a large set of markers before picking CD54 long before the phase 1,2,3 trials started. So it's a predicted effect and not just something that popped out of a retrospective exercise.
(OT) That is funny. But analogues of that visual method are common in discrete mathematics. For example, check out the so called q-analogs of equations involving factorials. Some of the q-analogs have turned out to be important in theoretical physics. One who has done much with that for many decades is George Andrews of Penn State U.
http://en.wikipedia.org/wiki/George_Andrews
When asked by the CBER Advisory Committee in February why CD54 was chosen as a potency marker for Provenge and why it should be correlated with long term survival, DNDN's Dr. Provost replied in a self deprecating manner that perhaps DNDN was just lucky.
Luck seldom figures in this sort of things - just lots of grunge work. Dr. Urdal discussed the antigen cassette technology and the role of CD54 at the Workshop on Tumor Vaccines in 1998. He'll likely talk about the CD54 data from the trials in the Cell & Gene Therapy Forum in Baltimore in Jan 2007.
http://www.fda.gov/CbER/minutes/tumor121098.pdf
"Now what I'd like to suggest in the next
couple of slides is an alternative way of actually
looking at some of the co-stimulatory markers and
their expression and picked CD-54 by means of
example, that it not only serves as a means by which
we can quantify dendritic cells that have been
induced to maturate in these cultures, but it's also
a marker that is associated with the potency and
capacity of these cells to actually interact with
naive T-cells."
Any trial where overall survival is the primary endpoint is a counter to Ocyan’s argument.
Well, ok. You seem a bit worked up and perhaps not thinking straight due to that. Let's try again. What did you think was my argument?
Yah. So where in those posts did you see your assumption of my assertion about some hypothetical 'FDA policy about “free” or implicit p-value for survival'?
You seem a bit worked up. If it makes you feel better, you can have the last word.
Dew - You should reread my posts 460 and 464 on iVillage.
There was no assertion in these posts of some "FDA policy about “free” or implicit p-value for survival".
Those posts highlighted the Coreg's pivotal trial where the survival benefit was discovered unexpectedly and how the FDA treated that case. That trial missed everything including the primary surrogate endpoint (p = .27). However, its survival endpoint had p value .03. The FDA reviewers themselves discussed whether the implicit p value for survival should have been .05. So, the eventual approval of the drug did give the appearance that the FDA had reset the p-value during their evaluation. That might or might not be true as we don't know if they had other data on hand in their consideration. But that was a good counter-example to the usual bear assertion that missing the primary endpoint was fatal even with an achievement in survival.
The discussion of surrogate endpoints in the 03/2005 ODAC mentioned in my previous post shows that, at least, among the FDA advisors, there is openness in not being strictly orthodox about dividing the p-value among such endpoints. Whether or not this will apply to D9901 is yet unclear. But a change in attitude among these statisticians can only be positive for the case of Provenge.
For someone not subscribing to your naugahide theory, these events can be seen as evidence against your perma-pessimism.
There is a different in looking at independent endpoints or independent interim looks for stopping and looking at endpoins that are surrogates of one another. In the former cases, yes allocating p value properly is a must. In the latter case, it is not necessary. I posted the below follow-up to Clarksterh's post on iVillage:
http://www1.investorvillage.com/smbd.asp?mb=971&mn=4570&pt=msg&mid=420053
"it's likely that patients will just choose to get Provenge earlier in the disease cycle, then get booster reinfusions later on."
Management clearly have this mind. The term "frontline" was mentioned a few times in the presentation.
Did anyone notice the statement on "numerous scientific presentations" in 4Q. Aside from the CD54 data and possibly a paper on the synergy of Provenge+chemo, what else could they present with the 9901+2a data?
P-11 patients are in early stage so rising PSA is the main sign of advancing disease. Secondary endpoints include metastatic disease and survival.
Yes. They likely only get these numbers on a periodical basis. That would seriously cut down the number of discrete events. So any estimation of p value would be susceptible to large variation in error.
In late 2005, an FDA official told me that all cell-based cancer vaccines will be evaluated by CBER with possible collaboration from CDER OO. This was forwarded to DM although he might have known that already.
Given the current structure of CBER, my guess for the chain of decision will start with Dr Raj Puri of the Tumor Vaccines Division and his boss, Dr. Celia Witten, Director of OCTGT. The outside panel, if convened, will likely be headed by Dr. Mule'. It's interesting to note that all three people were at the Provost's CD54 presentation. Judging by the questions and discussions in that meeting, the data were well received by the attending experts. Some of them will likely end up on the review panel.
In addition, Puri and Mule' were at the 1998 meeting where Dr. Urdal presented the Dendreon's antigen cassette process and the possibility of using CD54 for measuring potency. So if my guess is right, the BLA will be reviewed by people who are fairly familiar with what Dendreon has been up to from its inception. They should be able to give a fair assessment of the data knowing all the issues with a cell-based therapeutic such as variation between patients, delayed immune ramp-up, etc.
CBER vs CDER.
It would be a big surprise if CDER ends up reviewing the BLA instead of CBER. Thus far, all communication with the FDA was with CBER. This included the presentation of the CD54 data at the meeting of the CBER CTGT Advisory committee on how to measure potency of biologics. In the same presentation, Dr. Provost thanked certain FDA personnels in CBER for ongoing collaboration to clarify the measurement process. Lastly, the OCTGT has an entire division and a branch in it, both headed by Dr. Puri, dealing with tumor vaccines. It would be a slap on the face to the OCTGT if somehow the review of Provenge be wretched from them and given to the CDER Office of Oncology. There is no way that any FDA chief would allow such a fracas between CBER and CDER.
Also, Dr. Silvana Martino is no longer listed as a member of ODAC. She was the Acting Chair in March when Gemzar was reviewed. Maybe she left when their recommendation was overturned. The tide is turning.
http://www.fda.gov/cder/audiences/acspage/oncologicroster1.htm
From an IR email:
'The number of patients is based on "randomized" patients not simply signed up to see if they qualify...'
Fairly definite in this case.