Avid Bioservices Inc (CDMO)

[Protector]

md, you know I am a big believer in the fact that we will not have to wait for the complete end of PIII for the SUNRISE clinical trial.

However, after the annual meeting, I kind of tend to agree that the 1st look-in, depending on the #of events that will trigger it, may indeed be to early. Why, median, not average.

Here is the reasoning (and it is just a reasoning). 582 patients are enrolled and assigned to the treatment arms by a 1 to 1 randomizing. This means 1 goes to the SOC+Placebo arm, the next to the SOC+Bavi arm, and so on.

Historically the SOC only (or SOC+placebo which is considered the same) will event with a MEDIAN around 5 months (5.4 in our PII). So some after a months, others after more then 5 months. We saw in the historic data that some patients can survive 9+ months on Docetaxel alone. However, the 5.4 months is NOT an average and could be the MOS with most of the other control patients eventing before month 7. With a median you do not care about the face value of the numbers left and right of the middle element. This means that for the control arm we can assume that most patients will have evented within 7 months after start of treatment.

So if we assume (ASSUME) that PPHM aims for 80 event in the control arm (double of its PII control arm population) then 160 patients must be enrolled in the trial before this can happen assuming all Ctrl arm patients event within 7 months. 160 patients should have been enrolled in in JUL (at 0.3 patients per center per month). So in FEB 2015 all those Ctrl patients should have evented. And anyway, plenty of others have been enrolled in the control arm since JUN and some of them will event before 5 months so the 80 events should be possible by FEB 2015.

Then comes the question. Do they, based on that trigger, compare both tables/MOS OR is there also a number of events specified for the Bavi arm (that I don't know how they do that). Maybe they trigger based on Bavi arm events and compare then. Anyway, it doesn't matter to much.

The problem for Bavi SUNRISE to get stopped early at the 1st look in is that Bavi treated patients will live longer (should and we saw they did in PII) and therefore its table gets initially populated with the VERY sick patients for which Bavi came to late. For EACH such patient you need 1 that lives LONGER then the MOS of the control arm + 50%. And that costs time. The earlier you look the smaller the statistical chance this will be achieved.

But then how can ANY trial be stopped early. Because if the drug does well you have the above problem and if it performs poorly you won't beat the SOC sufficiently?

That is where the data monitoring committee comes in. They can see something else, the SIZE of the table. If in FEB you have 80 events in the control arm with a MOS of 5 months and you have only 10 events with a MOS of 2 months in your Bav arm table then THAT IS A GOOD thing.

But how can a MOS of 2 months be better then a MOS of 5 months. Well it can if the triggering isn't based on an EQUAL number of events in both tables. But that is what the median is designed for because you can almost NEVER pick a moment to snapshot where both tables will have an equal # of patients (unless all 582 event). Furthermore patients sometimes are censored and don't make the table.

Then if the median is used because of that why would a trial be stopped if Bavi has a MOS of 2 months compared to 5 of the control table.

The answer is survivors.

The Bavi MOS in this EXAMPLE is low because only our very sick patients evented but all others are alive and could therefore not populate the table with high values to the right. For the data monitoring commission that would mean AT LEAST 70 patients are still alive and ALL of them, being FEB 2015, can not do OTHERWISE then deliver a number to the RIGHT which is FOR SURE higher then the longest living patients in the most right position of the control arm. An that would be 7+ months were our 50% comes from. I would not be surprised PPHM has seen the detail of some historical control arms data and therefore has a good idea of the values to expect right from the control arm MOS and hence place a target of 7.5 months (50% improvement on median).

The above only holds because the trial is randomized 1 on 1. So it cannot be that if there were 80 events in the Ctrl arm there weren't at least 80 patients enrolled in the Bavi arm over the same period in such randomizing in case you would make patient #1 a Bavi arm patient.

You can see that such early stop would require bavituximab to perform EXTREMELY well BECAUSE between our above JUL 2014 and our ASSUMED FEB 2015 newly enrolled patients in the Bavi arm that are very sick will ALSO die and they will make the list longer but keep the median low. So our FEB 2015 assumptions of 10 entries in the table would have to include those ones too.

To put this in perspective, in FEB 2015 we should have enrolled more then 320 patients (again 0.3 used assuming no centres are added any-more). 320-160=160 of which 80 Ctrl and 80 Bavi.

The 80 extra Ctrl will not influence the results very much, statistically they'll perform in the same range with again very sick patients first but others from the first 80 will be added with longer survival times to the right. But the 80 extra Bavi patients will have there shares of events of very sick patients and they will not be added to the RIGH of the table.

So you see that the likeliness of SUNRISE being stopped at 1st look-in is possible but then Bavituximab must really be better even better the in PII and/or the control arm worse. This is however possible because in the PII the control arm patients received Bavituximab and Bavi patients placebo.


If SUNRISE would be fully enrolled in JUN 2015, then an Second Look In 9 months later could be successful. That would be MAR/APR 2016.
At that point ALL control arm patients should have evented and the BULK of the bavi arm patients will have had the chance to live 11 months (a little longer then the median of our PII).

We will have to see.