Cel Sci Corporation (CVM)

[Fosco1]

Dropout and condition for success
===================================

I see that the interest and questioning is raising because we feel that we are a few bits of information away to assess the full successfulness of this trial in advance with a pretty good certainty and this is exciting !

May I begin with what I am 100% certain (and I would like you please to believe it because I am 100% certain to be right and discussions around these topics have already been addressed thousands of times and solved).

Let me give a simple definition of dropout. In fact, the definition which is critical to us.
“A dropout is a patient who has been accounted as enrolled and whose outcome (dead or alive) has been lost and will be lost forever for the study at a point in time”

Again, sorry to insist, a patients rolled over by a 10 ton truck is not a dropout. He is an unfortunate man who only had an incurable disease and unfortunately died in a tragic manner.
He will be accounted as an Event in this trial if the sponsor of the trial is aware of his death because they are accounting “observed survival” and not “survival relative to the illness”. If the investigator is not aware of his death, he will be considered as a dropout.

Now a little bit on how they assess success in this trial (Primary endpoint)

1) Global efficacy is assessed on a binary way and needs significance
(Skip the following section if you don’t like details)

At end point and as well on regular basis during the IDMC meeting they will build two Kaplan-Meier survival curves spanning from year 0 (date of treatment) till a year corresponding to the timeline for the longest surviving patients. They are running from this the so-called Logrank test, which is a simple sum of gaps between the two curves and is aimed at assessing GLOBAL efficacy between the two curves. This is a binary test which basically says : “YES MK is efficient” or “NO MK is similar to control” and also gives a P which gives the significance of the results. Until P>0.05 then the TEST is not significant which means that even it says “YES MK Is efficient” and p=0.25 the credit to give to this assertion is that it is significant with 75% of probability which is insufficient they need like in all clinical trail a 95% of probability. So the IDMC will say “trial will continue” until they will see p<0.05.
Cel-sci calculated that significance should be reached (p<0.5) when when 298th event is reached which is the reason why we need to wait till 298 to happen. Significance could be reached sooner (let’s see what the IDMC says soon) if the two curves show a very neat difference (or similarity in case of futility).

2) Overall improvement of 10% in survival is required by the clinical trial between the two comparative arms

That’s a claim from Cel SCI . How they measure the 10% improvement is a bit of a statistical mystery to me (and not only me). However they might use Hazard ratio (an outcome of the Kaplan Meier analysis) which assess the % gap of deaths between both arms and is a global number for the whole curves and infer from the Hazard ratio the 10% overall survival improvement (“If we see 15% less events in the MK arm then it means 10% more survival”). Not sure.

NB 1: At no point there is a 3 years assessment. The Analysis is global for the two KM curves. Everyone who think there is something happening at Year3 survival milestone should forget about it. The analysis takes place at each IDMC meeting AND the last one, unblinded to Cel SCI will take place when 298th is reached. “This is an event based study”
NB 2: A patient who dropouts is not an “un-followed” patient. Because actually he is followed until he disappears. That’s what they call right-censoring in KM analysis. Their survival is tracked and accounted for in the KM analysis until they disappear. This data will be useful to assess global efficacy (as per point 1 above).

3) Max dropouts allowed

Generally speaking, max number of dropouts allowed in an event-based study is the number of dropouts which don’t prevent 298 to be assessed as events. In our case it is 928 – 298 = 630 patients, 63% : Obviously, if you lose more than 630 than you won’t be able to count your 298 events.

The reason they recruit so many patients is
a) To be able to complete the study in a timely manner
b) to reach statistical significance even if dropouts are significant

Now, when I wrote my SeekingAlpha article I mentioned in the comments to readers that they needed 784 evaluable patients (see Cel SCI scientific presentation) to assess enough power for the study. But truth is that to assess global efficacy, you don’t need 784, you just need 298 events. To assess the 10% improvement, you might need more and that might be the reason for the 784 but I can’t rule this in my certainties.
If 784 patients at least need to be followed (max dropout number = 144 patients) when 298th event comes, this means that there is a ceiling on the number of potential dropouts : 15,5 %.

And that’s why I didn’t use a Socs survival as bad as Cel SCI. Because with such bad SoCs survival as they published, this implies that the trial should have ended sooner OR that efficacy is mind blowing (miracle drug) OR that dropouts are well above 15.5%.

In conclusion
====================

If it is confirmed that 784 need to be followed from treatment to death, this means that we have a winner, with certainty, and I will personally bet the farm on it. Real SoC is likely better that the one they mentioned because , well that ‘s an average and patients in this CT are probably better treated with complete scale S-CRT, dropouts are contained under 15.5%.

This is a simple question to ask, and a simple answer to get. No need to ask for how many counted dropouts, this could be a blinded data for the sponsor.

Fosco