In making a forecast, and in our case, estimating the PFS rates for the control and treatment group, the quality and reliability of the estimates depend on good data and reasonable assumptions.
The quality of the data regarding patient enrollment dates and number of patients seems to be very reasonable. They have been collected over a period exceeding four years, and there appears to be no incentive for management to deceive us. I rate the quality of this data as good.
Could have it been better, yes. We did not receive precise dates for individual patient enrollment. We basically have 14 different lump sums of patient enrollment dates, along with the number of patients enrolled on these dates. We adjusted for this by assuming that the total number of these patients in each group were enrolled linearly from the previous enrollment announcement date. We then calculated an average enrollment date, which allowed us to calculate an average time in trial for these patients. How accurate is this assumption? Well, it depends on whether or not they were enrolled linearly over the time period. We have no way of knowing this, so we just live with it. I don't think it is much of a problem.
The next assumption we make is pretty critical. We assume 400 PFS events have taken place on or around November 1, 2012. We assume 400 because for obvious reasons. We know Celsion is not going to call for the trial data to be locked-down for analysis only to discover that the threshold level of 380 PFS patients had not been met. This would not only waste time but might enter biases in the trial that might be difficult to calculate.
Now, we test out a few scenarios. We first assume management's claim is correct and the control group has a 12 median month PFS rate. We run this rate across 350 patients over the length in time these patients have been in the trial. It yields a certain number of PFS events. Take that number and subtract it from 400 and we arrive at how many treatment patients have evented in the same time period. Given that number we work backwards to figure out the PFS rate for the treatment group.
I performed this scenario for over a dozen different PFS rates. Now, I will concentrate on three different control rates: 12, 15 and 18 month PFS rates for control. The following combinations of control and treatment PFS rates yielded 400 PFS events by November 1, 2012:
12 vs 36
15 vs 30.62
18 vs 26
It should be clear to everyone that if the control group PFS rate is 18 or less months that the trial will be a success. Provided are above assumptions are correct and reasonable.
A check I did on previous models and recently suggested that I do again by Trond is to run these different combinations of control group PFS rates and see what kind of results we obtain at the interim analysis. The difficult thing about the interim analysis is nailing down precisely when we obtained 219 PFS events. Summer claimed we had 600 patients enrolled around the fifth of August 2011.
Celsion announced on November 28th 2011 that we had achieved 219 PFS events and the enrollment at the time was 613 patients. Now, if we assume it takes the DMC about 10 weeks to do their review and get back to Celsion then that puts the 219 event around mid September 2011. Noting that patient enrollment over the past Summer 2011 was about 10-15 patients per month this might not be a bad estimate.
So running various PFS rates for control up to mid September 2011 for the interim analysis yielded the following combinations of PFS rates that yielded 219 PFS events:
12 vs 48
15 vs 36
18 vs 27
All of these rates would suggest that the trial will be a success. However, note that all of the combinations are different from the analysis of November 1, 2012, except for the 18 month rate for control.
Note: If all the assumptions are correct and reasonable then one might assume that the trial results might be more in line with 18 vs 26 than the other combinations. However, it is difficult to come to that conclusion. Why? Basically, we only have two known data point, interim and final results, and we are trying to shoe-horn two non-linear distributions to match those two points. That is a tough thing to do and is not very reliable. If we had 5 or 6 points I would be lot more comfortable with predicting a precise estimate of the two PFS rates.
I am left to conclude the following: If my assumptions are reasonable and appropriate then under any circumstance a control group PFS rate of 12 or 15 months will yield amazing trial results. A PFS rate of 18 months for control will yield good solid test results.
Where are we vulnerable? If the actual PFS events number 420 or more on November 1, 2012 and the control PFS rate is 18 months or more then the trial is in real jeopardy.
At this point I have seen enough numbers and simulations to last me for a month or so. I just want to see results. I am confident but always remember to only invest what you can afford to lose. This is a for real binary event. But I have to think the odds are with us.
Davis let me 1st say you are the one guy I like reading on this message board....you got me into this stock and I'm riding this baby all the way to the end with Feb calls (although this wait is killing me).
Any after kissing up to you...I thought I'd direct you to a study done from January 2001-March 2007....google "Local Tumor Progression After Radiofrequency Ablation of Liver Tumors: Analysis of Morphologic Pattern and Site of Recurrence" It comes right up. It's an easy 13 page read / interesting. In that study 1,756 patients with 2,431 HCC's received radiofrequency ablation alone...the time to local tumor progression on HCC ranged from 90-1,347 days with a median of 367 days (i.e. 12 months). I'm thinking this may be the source of management's assumption that the control arm should have a 12 month progression time period.
I sure hope we are reading this stock right on being long! Good luck to all....and Davis let me know what stock you pick next...you need to start an investment blog and I'll buy you a nice bottle of wine whenever a pick comes home........
Sentiment: Strong Buy
no, no, and no. This is a study where they ONLY looked at patients with local recurrence. We certainly can't assume that all patients will have LR. And the expected 12 month median for HEAT is for *any* of several kinds of events.
Sentiment: Strong Buy
davis. you have matured into the best poster on the board which is mostly useless lately. all of our previous longs and experts in the science appear to have run for the hills. biopharma mostly sold and wasted a chunk of the rest on expired options. kid and odaat haven't been heard from. I really wouldn't be surprised if the trial fails if they come back and post that they sold everything at $9+. I actually imagine odaat got committed and is in a padded cell right now with no internet access. He has a bit of a crazy personality. Posting that the DMC would be on 1/4 encouraging people to buy Jan options and then saying he hopes results come out after just to burn them. I really don't like anyone on a message board posting that they have inside information.
I'm ready for this to be over however it plays out. It just seems logical to me that the addition of high doses of localized chemo at the tumor site should make a difference. I always thought the critical thing with this treatment was safety which could kill us but I think the interim pretty much closed the door on that.
Sentiment: Strong Buy
So glad to read that you have missed me. I have been lurking since my return from Antarctica (best trip ever). I see no reason to jump in.
Lots of garbage on the board right now. Frankly the SP going into the data release is pretty much irrelevant. who cares! What is important is what it will be AFTER data release.
I feel no compulsion to try to convince anyone one way or the tother . All that needs to be said about science and trials has been said over and over again. Some of it has been good, some of it has been bad. Nothing new to add.
By the end of next week we will have results to discuss. Now that should be interesting. See you then.
Davis , when you first showed up with your fancy models and stuff many here did not know how to take you . I said that you were just using an if then model , if 12 months for control , then 52 for the active. Over time you have listened to the feedback and revised your model but the same truth still comes out . If the company and its investigators are even near right on their PFS assumption then the Heat trial will be a huge success ! Now that you are buying friends with fancy liquor I am sure you will fit in here just fine . Data comes out Jan 2014 right ?
Sentiment: Strong Buy
Davis, I would not discount the two timeframes converging on 18 v 26/27.
1) that fits the projected median total time of 22-23 months or so.
2) a very narrow miss at interim fits with Kid's 19v27 post.
3) Several new studies that Sia and others have found make me rethink "12-months-as-gospel." One Chinese one-center study had 95 RFA-only, up to 7cm tumors with a PFS calculated around 25-26 months. Ouch!
4) Most important, having the two timeframes converge really makes my itch go away. If a model "fits" one timeframe but is out of whack for the other, there obviously is something wrong with the assumption in the model.
I'm still bullish, 18v27 would suffice for success, but especially here I'm scaling back any assumptions. My official estimate for your contest is still 14v24ish. :-)
"One Chinese one-center study had 95 RFA-only, up to 7cm tumors with a PFS calculated around 25-26 months."
BOMBSHELL. Management's 12 month control estimate is out the window bro. What happens when the models are run with a control PFS average of 25-26 months? The trial undoubtedly fails. This is what the PEOPLE need to be thinking about. First distant mets, now RFA making huge advances? Any other trial killing variable out there? And you wonder why management is dragging their feet on the data? Time to bail bros! TRUTH.
Sentiment: Strong Sell
I hear what you are saying and that is why I posted the results after I analyzed and re-analyzed the results. I just wish we had more than two data points. And yes it is a bit more consistent with the 22 month total rate.
I agree that 18 vs 26 is still a very good result but it leaves us with a little less wiggle room. The way we are calculating PFS means those control patients can still be eventing up until the 21st month scan and still counted as in the 18 month PFS rate time period. That is some cushion. Other studies have tighter reporting windows. So we must remember to be equating apples to apples. We ave room for error.
These results are going to provide a lot of answers, not just to modelers but to liver cancer patients and the whole notion of effective targeted chemotherapy can be. Exciting times for everyone.
I am so way past ready for these results.
Cheers and Glty.
sdtrond- Do you happen to know the name of the study you are referencing. Generally, I agree that many studies have low local recurrence rates but when you add in distant recurrence (and these are studies with smaller tumors) you are often near the 12 month time frame. Assuming you also include patients that fail twice, deaths, extrahepatic etc....do you still think it might be a real stretch to assume 12 months. Its funny because when I listened to one of the scientific/investor day presentations by one of the docs he states that most patients will have a recurrence in the 8 to 12 month time frame.....
Allow me to join the chorus of well wishers and among those appreciative of your efforts and sharing with the CLSN investor community
Your posts and tweets are among the 1st I read
To our greater success
Sentiment: Strong Buy
Awesome work DCup. I've enjoyed your posts but everybody should remember that the length of the trial is in-line with what CLSN management modeled for a successful outcome and what the trial was designed for. I'm willing to wager they have modeled a safety factor in to the trial as well - in other words 12 months for the control group . Maybe this is why the insiders are buying shares and not selling shares. Confidence not con-men. Think about it. It's not that difficult. If two shorts have one brain cell a piece, you can get together and see the light with only two brain cells. Good luck to longs. Shorts thanks in advance for the pop.
Sdtrond, nothing comparable out there? What makes Yin first-treated not comparable? It's exactly the same size tumors and treatment naive? (The only thing that could be negative that I know of is that the sample is small = 58 first-treated.) And the CMO of CLSN thinks it's the most comparable? Yin first-treated doesn't calculate PFS but it still hits one in the face (at least me) as having less than a 9 month PFS (especially if translated into a study using CT rather than US) and certainly 11 months is the top limit PFS that I can see anyone figuring from Yin et al even if staying with the US search with CT validation that Yin uses? To me, even if RFA has improved, the translation of US search to CT search still likely leaves Yin with an 11 month or less PFS?