Replies to post #131647 on NorthWest Biotherapeutics Inc (NWBO)

[sentiment_stocks]

Thank you for that explanation CDRD.

So to be precise,
the only patients that count towards an actual median OS number are those that have actually died?

If this is the case, than Larry Smith was very correct in his statement.

I'm just not clear if they count when calculating the median OS for the trial - since they hadn't actually died by the end.

And that makes the most sense to me, but I am just seek confirmation of this.

However, I am very clear that the censored patients will count in other, equally or more important, data.

[Tadasana]

cdrd, shouldn't there be two medians? one for control (110 patients median at 55th patient death then if there are alive patients in control group they will be censored at time trial is unblinded) for vaccine (220 patients median at 110 patient death with rest alive patients censored) median should be 165 patient for blended in a trial of 331, and only patients that die to the left of the median will move the median left other wise if to the right of median will have no effect? That is why they wait for the long tail to appear for the 25-30% of patients that don't die and are alive at the long tail. This is when the curves separate for the vaccinated versus control arms, but ours will be confounded slightly due to crossover affect but still the PFS numbers will not be confounded and it will clearly show the long tail of survivors in the vaccinated arm the are alive and PFS. imho

[eagle8]

Thanks CDRD.

And the events yet to occur will only move the needle to the right-hand side, because OS of the 100 is already long and they all will come at the right side of the excisting median, the one for 221 patients.
100 (or 98) O.S events left to move the median needle more and more in the right direction.
Maybe they can extrapolate or give the 100, still alive, a fixed place in the curve at were they are now. Both would give a big move.


My $0.02

GLTU

[
CDRD Sunday, 08/27/17 07:55:34 PM
Re: sentiment_stocks post# 131637
Post #
131647
of 131659 Go
I think there is an easier explanation for the mOS than has been offered to date.

If NWBO stops the trial at precisely 233 OS events, the median will be the 117th patient (when all the deceased patients are rank ordered by how long they lived). If NWBO waits unitl 234 OS events, then the median will be the average of the 117th patient and the 118th patient. If they wait for 235 OS events, the median will be the 118th patient, and so on.

So, Senti and others are correct that the more OS events that occur the more the mOS shifts to the next patient---these differences may be small but that is how medians work in this case.
/quote]

[exwannabe]

If NWBO stops the trial at precisely 233 OS events, the median will be the 117th patient (when all the deceased patients are rank ordered by how long they lived). If NWBO waits unitl 234 OS events, then the median will be the average of the 117th patient and the 118th patient. If they wait for 235 OS events, the median will be the 118th patient, and so on.

That is wrong.

Once they have the shortest 167 events that is the median. But prior to that being known, the reported number is the KM ESTIMATE of where that final median would be. Specifically, it is where the KM curve crosses 50%.

As an estimate, it can shift around. But it doe not simply keep increases when time.

The problem with using what you are suggesting is that it is always time biased. The reported number will be an less than the real number, and rise up to it over time. Kind of a useless metric.

A picture is worth 1000 words, so here is an example. Scroll down about half way and look for the PFS data. The median for treatment arm is 10,3 months. Now open figure 1 and expand it (nice slider at the top right makes it easy).

You can see the median is where the curve drops from just over 50% to under with the first of those 2 events occur almost at the same time at 10+ months.

There are clearly only 2 further events. So the number of events after the ESTIMATED median was only 3 (with 75 or so earlier). This is clearly not even close to the ordinary median obtained by ignoring the censored data.