Hi meirluc.
I suspect Ex is broadly right on this from his look at the curves.
I tend to come at this from the other end. Which is that I don't think that alphapuppy's model predicting such an amazing p value in favor of DCVax treatment on PFS can be right, because I don't think the p value can be right.
If it was, then they would most probably have received a 'halt for efficacy' rec from the DMC back in 2015, and we would have AA by now.
So I just assume that the modelling has limited validity for one reason or another.
It's just a model, in the same way that all those Covid models were.
When the curve turns into a completely flat straight line, I just think 'I don't know what this really means and does it mean anything?'
A bit like the infamous Optune high compliance curve that turned into a flat line, and which didn't really mean anything.
I'll confess to not really liking KM curves, as I find them difficult to interpret fully. Though they can tell you important stuff sometimes. Crossing survival curves is the prime example for me. That is seen in many ICI trials where mortality on the treatment arm is higher than on the control arm, before things switch around with further time duration. You can see that at a glance.
Plus crossing curves also tells you you don't have proportional hazards.
In the case of Alphapuppy's video, I didn't even scrutinise the curves too closely, because I find them difficult to interpret, and I've already assumed they must have a large inaccuracy factor.
And on the IMUC trial, there were only 42 control patients.
To what extent can 42 patients from another trial in another time and another place actually inform you how DCVax control patients will PFS perform?
Not very much, I think.
And as with OS, measuring PFS by median in our trial is not a good idea (if you want to best elucidate the benefit). It doesn't capture all those long term survivors, who not only survive, but do so without progressing (just like Brad). So if NWBO are going to use PFS in the final analysis, they are likely to employ restricted mean progression-free time.
It means a lot to a patient if you don't just survive, but you do it without any recurrence of disease. You are not only living longer, you are also living better. Most of this group are undoubtedly meth.
We know that unmeth patients comparatively, didn't do very well in the trial, so I doubt there will be a huge difference in PFS between control and treatment for that group, frankly. Certainly not by a measure of the median.
Alphapuppy has acknowledged that his model is based on scant input data, and comes with big assumptions.
And it's not my intention to be critical for the sake of it.
It's very welcome.
But personally, I don't read too much into it, in terms of predicting the PFS delta in our trial.
AI
