Replies to post #252610 on Avid Bioservices Inc (CDMO)

[JamesGMS]

Exwannabe, not so sure you're correct.

The looks are at 33% and 50% of that 466.

Golfho:

Quote:That implies that there is, in my view a 28% probability that the trial could be stopped at the first look-in.


Nope.

The trial can not be stat sig at the first look as there is no alpha assigned (if we believe SK). So one can not possibly assign a percent chance of being stopped for such.

Can't remember which meeting - but quite some time ago Dr. Garnick was asked about a possible first look-in halt. As one would expect he replied that he thought it was highly unlikely - but then he went on to say that it could be possible if the results were "spectacular."

Taking a look at tradero's recently posted projections - assuming they bear some semblance to the ongoing reality in the trial - one could quite plausibly refer to the clearly diverging survival numbers between the control and treatment arms as at least bordering on being "spectacular."

So it would appear that Golfho was correct, and a possible {1 in 4 chance by his reckoning} first look-in halt is clearly still on the table.

James

[tradero]

I think there are good reasons to be optmistic.

Thanks EX for your reply.

The trial can not be stat sig at the first look as there is no alpha assigned (if we believe SK). So one can not possibly assign a percent chance of being stopped for such.

I agree that the 1st look-in may not be stat. significant.
However, we have good news in front of our veryy eyes IMHO.
Since you talk about numbers, let me repeat what may have been lost in the too many words I used in my last unreadable post.
Just from the info we have right now, we can have big hopes that Sunrise trial is doing great.

Let’s see:
1. Do you agree that the Placebo Arm in Sunrise will behave pretty much like the Herbst's Trial (Doce + Placebo)? We should bear in mind that the patient population and number of patients are very similar in both trials.

2. Do you also agree that the enrolment model based on the dates when Sunrise Centers joined the trial can give a “fair” enough enrolment curve? Well, I know this is debatable but I don’t think that reality differs much from this:

A hockey-stick type of curve and on the conservative side regarding enrolment, and let’s remember we are playing conservative for enrolment (and therefore for the eventing part).

With these two assumptions in mind and some simple math, I believe we can estimate the number of events that may have taken place for the Placebo Arm at a particular moment. We may be guessing the statistical results for Bavi, but we do have stats. results from the Placebo Arm.
And what I tried to show in that post is that by the end of this month the numbers say 119 events, and 130 events by the end of March, both for the Placebo Arm.

Even if the model I used may not be too accurate, I certainly think that there are already more than 100 events for the Placebo Arm (if it behaves like the Placebo Arm in the Herbst' trial)

If 1st look-in has not happened yet (fact, LOL)… In my opinion this means that the events coming from the Bavi Side are not half of the numbers from the Placebo Arm (around 150 events needed for 1st look-in?).

For me this a good reason to have high hopes for the Sunrise trials.

[Protector]

exwannabe, I think that calculation needs a slight change. This is how I see it.

PPHM is expected to get 33% (1st look-in), 50% (2nd look-in) and 80% (final unblinding) of the # of patients needed for calculating the unblinding 80% as specified in the trials protocol between PPHM/FDA.

We know from the PPHM PR that when 90% of the 582 patients were enrolled (in DEC 2015) the number of REQUIRED patients to unblind were enrolled already. So that number, required by the FDA, is MAXIMUM 582*.90=524 (rounded up).

That is a MAXIMUM because PPHM did NOT say they needed the FULL 90%. They just said that with 90% enrolled all needed patients to unblind were enrolled.

33% of 524 = 173 events for 1st look-in
50% of 524 = 262 events for 2nd look-in
80% of 524 = 420 events for total unblinding

So I think the 466 number is incorrect if the reasoning above is correct, which I think it is.

Question: How can they not have assigned an alpha? The FDA, if not specified otherwise, uses 0.02 as the alpha. So p<0.02 would be stat. sig. In math-stats it is often 0.05.

When did SK say there was no alpha? If that would be correct the 2nd look-in cannot check for efficacy because it would not be possible to see if that result was statistical significant.

[golfho]

Just catching up on all of the posts…

The trial can not be stat sig at the first look as there is no alpha assigned

Admittedly I know little about clinical statistics. So I do not know that much about alpha other than what a quick Google search provides like this:

http://blog.minitab.com/blog/michelle-paret/alphas-p-values-confidence-intervals-oh-my

Or this:

https://en.wikipedia.org/wiki/Clinical_significance


However from the above definitions & explanations it appears to me that alpha, p-value and come to think about it…All of the statistical parameters that the trial was designed with, have not been revealed. My recollection is that the only statement management has made on the subject was that the trial was powered to show a 2 month improvement over SOC…Or something like that.

So if you please, could you explain in more detail what your understanding is and what you are trying to convey.

Regards
golfho