Celsion Corp. Message Board
-
Seeking Alpha Model Analyzed...
Today's Seeking Alpha article "Celsion's Phase III Blockbuster Data Revealed And It's Been Right Underneath Your Nose," takes a very similar approach to my original model, which I had posted here a couple of months back.
Some have asked me to analyze it so I will.
The model is strikingly similar to mine in its approach. It does not attempt to divine how well the ThermoDox patients should do based on the science or scant Clinical evidence we have from our Phase I trial. Rather, it takes the total known PFS events and then subtracts what the expected number of events from the control group should be, based on a significant amount of data we have about our RFA patients, enrollment data and time in trial, and then arrives at the number of PFS events from the treatment group. From here it a very simple calculation to determine how well the Treatment group is doing in comparison to the Control group. After running the numbers he concludes the Treatment group is exceeding everyone's expectations.
Let's look at the models side by side.
1. He calculates a 10.5 median monthly PFS rate for the Control group. I use a 12 month median.
2. His enrollment data and time in trial data match almost perfectly with mine from what I have seen so far.
3. His Total median PFS rate is 25.6 months. I arrived at 25 months.
4. He does not calculate a PFS rate for the Thermodox +RFA group because he claims less than half of the patients have evented, so this is not possible. Uh, this is not right. He calculates there have been 123 events from the ThermoDox patients. Well, given that we know the time in trial for all the Treatment patients it is just a matter of running a number of different PFS rates over the times in trial to arrive at a PFS rate that will yield 123 events. He did not do this. Oversight, perhaps.
5. He claims that the ThermoDox + RFA patients will show an improvement over the control patients in the neighborhood of 144 percent. There is a serious error in his calculation.
He arrives at his figure by noting that the Total PFS rate is 25.6 months. He also calculates, if you recall, that the control arm has a 10.5 PFS rate. He then performs the following calculations:
25.6/10.5 = 2.44. In other word, according to him, it takes the Treatment arm 2.44 times longer to event than Control arm. That is a 144 percent improvement. This is a serious error.
The error stems from the fact that he is comparing the Total PFS rate to the Control PFS rate. He should be comparing the Treatment PFS rate to the Control PFS rate.
The relationship looks more like this (Not adjusting for the non-linearity of the patient enrollment data and the non-linear PFS distributions):
Total PFS rate = .5 x PFS rate of Control + .5 x PFS rate of Treatment, hence
25.6 = .5 x 10.5 + .5 x PFS rate of Treatment.
20.35 = .5 x PFS rate of Treatment
40.7 = PFS rate of the Treatment group.
Now, comparing Treatment to Control yeilds:
40.7/10.5 = 3.88, that is the treatment group takes 3.88 times longer to event.
That is a 288 percent improvement for the Treatment group over the Control group.
Conclusion: I agree with his analysis in the main. Given the historical data on seriously ill liver cancer patients with tumors in the 3-7cm range and their tendency to even in under 12 months, ThermoDox must be working exceeding well. His model is roughly 10.5 vs 40.7
Nice, very nice.Sentiment: Strong Buy
-
excellent, excellente
-
Straight forward and thorough analysis. The numbers are staring you in the face. The results could be released at any moment, and shorts, how difficult do you think it will be to try and cover when the whole world is buying?
Sentiment: Strong Buy
- 1 Reply to daviscupper
-
Dcups,
I suggest you up your Prozac dosage to 40.
You are obsessing. Over the last week you have been posting from 6 AM to 2 AM. Incessantly.
Your tone indicates you are losing control.
Dr. Kadglish's "relax" appears insufficient.
A flying monkey
-
The model is very persuasive. What didn't you like about it shorts? You seem to avoid this post like it is kryptonite or something. Cheers.
Sentiment: Strong Buy
-
Excellent post$$$$$$$
PrinceSentiment: Strong Buy
-
Bump
Sentiment: Strong Buy
-
Can we be 100% sure that "a treatment failure" means Pfs event at t0.
-
Dcups maybe I'm not understanding something, but his model has probabilities for local and distant progressions. However, they aren't mutually exclusive..you can have a distant progression prior to a local. From my look, I don't see he accounted for that. Seems like that would slow event rate.
- 2 Replies to chrismarkus1
-
Nope, I most certainly accounted for that. Check the excel spreadsheet within the article then check my math. As an example, at the 24 month mark I used a 50% chance of LTP and 50% chance of IDR for an overall 75% chance of PFS. 75% instead of 100% accounts for a 25% crossover. Each data point is formula driven.
-
That is getting into the science and others are more informed about the overlap of distant and local mets. I seem to recall in the Yin study or maybe it was the Beber study that the overlap was slight. Maybe he accounted for the slight overlap in his probabilities.
At any rate it really does not detract from the main thrust of his model and argument. If the control PFS is anywhere near 12 PFS then the results are going to be "off-the-hook."
Cheers and GLTY.Sentiment: Strong Buy
-
Is the bear farting his brains out, again. I wish he would stop and take a few moments to share his brilliance and insights with us. Especially now when we have a critical binary event coming up.
Sentiment: Strong Buy
- 1 Reply to daviscupper
-
davis- let me know how I insulted you so I can properly apologize. I really didn't mean to. I don't see how the considerations we were discussing have been built into your model, but maybe I just missed the most recent version of it. What's your estimate now? Is it still like 200% improvement?
Sentiment: Strong Buy
-
daviscupper,
from an increasingly nervous long - many thanks for your potentially enlightening analysis. curious about one point: how are you able to determine the point in time when each of the 700 subjects were enrolled in order to know how long each spent as a subject of the study? i know that the november end point is public, but wouldn't you need the starting points in order to measure # of months of PFS? is this information available somewhere or are you able to extrapolate it somehow?
thanks againSentiment: Strong Buy
- 4 Replies to machinious
-
I would not thank anyone until the data is out. These posts like the results are a foregone conclusion are making me nervous. The chickens have yet to hatch so let's stop counting them.
Sentiment: Hold
-
Great question.
If you search back and locate my model (It is four post long all in the same thread.) you will note there are, I believe, 20 enrollment dates with various numbers of patients reported on those dates. I realized we were only getting updates in batches. I knew some patients had probably been recruited shortly after the previous announcement of enrolled patients and some were probably just recruited right before the current announcement. I therefore assumed all the most recently announced patients were enrolled at the midpoint date between the last announcement and the current announcement. You can only do the best you can do with the data you have.
By the way, there is no need to be intimidated by my model. It is all pretty straight forward. I made the math very accessible without jeopardizing too much accuracy. Check it out. You will have to hunt because I can be a prolific poster. Cheers.Sentiment: Strong Buy
-
dude, no reason for your nervousness to be increasing. nothing has changed. we have always known this is not a sure thing-- treat it as 50/50 for your investment purposes. so only invest what you are prepared to lose in a coin toss. but to answer your question, we can extrapolate the rate of enrollment from company PRs over the years saying how many patients they had enrolled by that date.
-
There were about 10 different PRs over the course of 2008-2012 with enrollment updates.
Available from the company website, or from any of dcups or alex's analyses.Sentiment: Buy
-
madipbart,
Thanks for the kind words. It won't be long now.Sentiment: Strong Buy
- View More Messages
-