Jammy's point, I think, is that what really counts in this trial is the P value, not measure of degree of separation. The P value is the same regardless of using median, H/R or landmark as the effect measure.
The P value is presumably log rank. This does not assume any sort of proportional hazard. [I assume "bog standard K/M" was a typo]. The K/M curve itself is the basic starting point for every stat that has been discussed.
While I'm happy for you to interpret what you think is his point, I'd also like him to respond to the issues I raised in response.
Why do you think his reference to 'bog standard K/M' was a typo?
Strange sort of typo. Doesn't read like a typo. Why do you think it is a typo?
I would somewhat
agree that a P value is what is often quoted. But the P value has to be attached to a particular stat.
What stat do you suggest they might use?
I sometimes see a percentage relative reduction in risk
in toplines (as your example). But this would be derived from hazard ratio.
And you don't use a singular hazard ratio, unless you have proportional hazards. And anyway if you use a hazard ratio on OS, what is your comparator. Is it control who didn't cross, control who crossed, a blend of control who crossed and those who didn't cross, controls who crossed after pseudo-progression, controls who didn't cross after pseudo-progression, controls who crossed after actual progression, controls who didn't cross after actual progression etc etc...
I guess quoting median OS/PFS can have a P value attached, and can be taken straight from the curves and doesn't use a hazard ratio.
But we don't want to use medians for reasons oft-stated.
So I will be interested to see what they do use.
And imagine they will use mean
survival and milestone
I accept that your example topline is not unrepresentative.
employ percentage relative risk and attaches a P value.
But like I say, quoting a relative risk comes from a hazard ratio, and the hazard ratio presumably comes from Cox regression analysis.
But more generally, who says
your endpoint has been met, allowing you to put that out in topline. Does it require your DMC to reach that conclusion? That would mean your DMC being onboard with any SAP revisions.
I do see some toplines that don't include any
stats, but just just blithely state that an endpoint has been met, or efficacy has been demonstrated, but failing to give an evidential stat.
I guess we will see in due course.