Cel Sci Corporation (CVM)

[Fosco1]

Hi all, first post here. Thanks for the (semi) relevant and (semi) knowledgeble posts.

I am very familiar with this trial and am very familiar with trial statistics.

Hi, good news ! Good to have knowledgeable people on board !

I do like foscos model, but my biggest concern is that it does not seem to show the delayed benefit of MK, rather it appears to benifet survival immediately.

My model assumes a constant survival improvement VS SEER Baseline, which does not mean the benefit is immediate. The benefit is calculated for a given date in time, therefore applies the delay between time of treatment and this date.

For instance in latest sheet, which was frozen in time at the 298th event (set as of 20th April 2020), latest enrolled patients from Sept 2016 had 3.6 years since treatment and average for all patients about 5years. If delays occurs for instance at year 2, then is it (or not) a good assumption to assume OSI of 24.5% for instance on those Sept 2016 patients as well ? May be OSI would be better for earlier patients and a bit under for the latest patient (progressive divergence of survival curves) ? That would be an different model which would imply progressive stronger benefit as time passes, but this would complexify it, for a delayed benefit that is still being an hypothesis, not a fact (everyone being blinded).

This would not jive with the thesis that the death rate/MK improvement was not clear ~5 years ago when more patients were thought to be needed.

Not sure what you mean. CEl SCI was blinded, so more patients were required because death rate was too low. There was no way to say whether MK was the cause or SOC was surviving stronger than expected. More patients were needed to increase the sample for a faster outcome

I think it is more likely, like others have said, that the MK and SOC curves were nearly identical until 2015-2016 and since then have begun to seperate.

Again, Cel SCI spoke of slow death rates, not of lack of MK benefit.
But OK let’s then imagine CEL SCI « saw » (but how, we do not know as, they were blinded) lack of benefit in early 2016. Average year since treatment 1,3 year. Likely number of events below 90. Delay might have occured afterwards and CEL SCI "saw" benefits back in 2017 with more than 133 events, that was 1,8 year in average for each patiennt. They delay is more than 1 year and less than 2 years. My model fits with delay.

I think if fosco's analysis took this into account you would see the osi drop closer to 10%.

I do not see how you would obtain this result, unless you draw a new SOC baseline curve that has higher survival than the usual SOC baseline in both the years before any multikine improvement ( before year 2) and after as well, which would be a strange way of doing things.
Everyone that will use a SOC baseline higher or much higher than SEER in every year (no only the early years) will prove a lower benefit.

SOC baseline survival is a fundamental assumption of this trial (with number of dropouts) and to prove or disprove trial success, one has to produce a SOC baseline and the following rationale that supports it.

Like I said , there is the so called « clinical trial effect » that could affect the survival of this population VS SEER data, effect that if we assume is above +50% of SEER's SOC in every year (and in particular after year 3) would jeopardise the positive conclusion of this trial.

I am also concerned with statistical significance as we approach 10% from above. With a 3.5% yearly dropout, statistical significance actually become a serious issue even around 8% or 9% osi.

Indeed, around 10% significance is in the danger zone.
All success / failure will depend on your SOC baseline assumption, therefore you can be concerned if you are convinced that your SOC baseline is true.

Would anyone, maybe sushi or fosco, be able to comment on the timing of the divergence of the soc and MK arms? It's effect on osi and how it is or isn't accounted for in foscos model?

In my opinion, if divergence occurs after year 3, it should not affect the model, unless one choses a very « special » SOC curves that fits a pessimistic scenario. It could affect though the p value calculated from Logrank as p derives from a comparision of two survival curves (SOC and Treatment), and the Hazard ratio if calculated trhough the proportional hazard method (because it would not be proportional). Now, if survival is above 20% (after year 3), i have not done a p calculation, but I still believe p would be below 5% between both survival curves that run from year 0 till year 9

Fosco