Replies to post #130031 on Biotech Values

[iwfal]

What are you referring to when you reference trigger here?

Not sure what in particular you are asking - so I may be overly rudimentary in my answer. Most time-to-event trials are ended when a certain number of events are hit and that number of events is normally a fraction of the total enrollment. For cancer OS it is generally around 75% - but I've seen low 70's and high 70's. (Note that for other endpoints it can be substantially different - e.g. PFS is often in the 60's because of the higher rate of LTFU)

That number sounds low to me given that enrollment was completed over 18 months ago and we have still yet to reach the pre-specified number of events (deaths)

You are at the tail of the curve - events slow down. For example, if you had a trigger of 95% of all patients must have died then the trigger wouldn't be hit for many more years.

Also, how do complete responses impact measurements of overall survival? ARRY had previously disclosed a complete response in the Phase 1 combo trial of selumetinib in BRAF melanoma patients (no guarantees it wasn't in the control arm of course).

If you think mapping historical data onto your trial is difficult ... it ain't nothin' compared to trying to map Response data onto survival.

I think historical range of survival in this patient population is the 8 months that ARRY management referred to.

Typically the range of survivals seen in historical comparisons of similar patients seen in other trials seems to vary over about a factor of 2 - i.e. if you find 5 trials the lowest and the highest median survivals often seem to be somewhere around a factor of 2 different. Sometimes more, sometimes less. Perhaps this patient pop will have less historical uncertainty - but you'd need to find the historicals to know. I, personally, don't even look until the trial is running long enough to indicate at least 50% over the 'nominal' median. ('nominal' being what the longs and the company talk about)

[iwfal]

BTW - A good rule of thumb for cancer OS is that the trigger will be 'expected' at:

End-of-enrollment + (2*Expected Median Survival)*1.1

This works because:

a) Most trials are back end loaded for enrollment

b) Survival curves are generally exponential - so going from no one evented to 50% evented (median) takes the same time as 1/2 of the remaining eventing (which is 75% evented).

c) The 1.1 factor accounts for several things like - the trial isn't completely back end loaded and there will be some lost to followup and there is some delay in reporting.

NOTE: Similar ROTs can be developed for PFS etc - but the factors will be different because the trigger is different and the LTFU is different and ... .