Avid Bioservices Inc (CDMO)

[golfho]

How I got there…

This in response to exwannabe’s & tradero’s posts of a few days ago.

Ex, there are several questions that you raised in your post that I would like to discuss further however, I’m not sure that I fully understand the details of your questions.

“there looks to be a mechanical error when he gathers the table describing an event rate based a similar trial, then adjusting for median.”

“Look at all the 20 month survival values. On the pattern trial GH uses, the 20month survival is about 30%, yet GH's is 4% for the placebo arm. These numbers do have slightly different meanings due to drop outs, but that effect will be minimal.”

The first thing I would like to explain in more detail is the evolution of my projections and its current goal.

About mid year 2014 management stated that they would start a PIII NSCLC trial by EOY. Much discussion ensued on this board. Lo and behold in December 2014 SUNRISE trial began…with one site active. In January 2015 2 sites were added. In February 6 sites were added, then in March only 2 more sites were added. At that time I became concerned about the company’s ability to achieve the stated goal of completing enrollment by 12/2016 because of anemic site participation.

At that time I started a spreadsheet that tracked site openings and locations. Then in rapid secession April, 11 sites, May, 38 sites, June, 30 sites, July, 29 sites, August, 12 sites. 120 sites opened in a period of 5 months. Ahhhhh…A small sigh of relief…But not for long.

The company historically has not met it’s time table for enrollment on its previous clinical trials, so I turned my attention to enrollment projections. I arbitrarily selected several scenarios, a patient every 3 months, a patient every 4 months, a hybrid model of a patient enrolled within 2 months after site activation and then every 4 months thereafter and finally a model based on the known enrollment rate for the PII trial ~ a patient ever 6 months. Over the following year+ additional sites were added, other posters made their projections and we all waited for things to mature.

Around mid year 2015 it became apparent that the more optimistic projections were to optimistic. With continued assurances from management that SUNRISE was on track the pessimistic projection using the historic enrollment rate was also removed. That left the enrollment rate of one patient per site every 4ish months still in play. So I expanded the projection from monthly to weekly. The projection that fits best with the numbers provided by management turn out to be ~ 4.5 to 4.75 months (137.6 to 144.6 days).

Around the same mid year time frame the subject of timing of the first and second look-ins arose on this board. I and several others played with some numbers. As a matter of fact you provided some guidance on the subject by pointing out that the look-ins were based on “33% and 50% of a predetermined number of events” and not on the total number of enrolled patients.

So now the question of; how do you model the events arose. I thought about it for some time. How do you model the MOS distribution? How do you account for the long tail of the KM plot for Bavituximab? How do you deal with the early events leading up to the first look-in? Will there be sufficient separation at the second look-in to compel the IDMC to halt the trial at the second look-in?

After several weeks of scratching my head and playing with numbers I had an epiphany. When I laid out a series of numbers ranging from 1 month to 20 months regardless of the number of numbers I used the median and the mean came out to be nearly identical. What will skew the results in one direction or the other would be 1- several large numbers on the high side or 2- A large series of very small numbers.

Generally, engineers work toward the goal of “Near perfection” but in reality we always arrive at the point of “Good enough” I made the decision that in this particular case I could use mean (Average) as a substitute for median. The error introduced would represent a week or two in one direction or the other. No big thing. I then selected a number of probable MOS outcomes and populated my spreadsheet. As time went by, the predicted low number MOS’s fell out.

Bare in mind that my goal was NOT to predict or project a final MOS number but to project a time frame for the look-ins and secondarily to estimate roughly the possible number of events per group. That secondary goal is what prompted me to write in a previous post #243665 in November:

“Now for my interpretation of certain words:
Possibility: I use this term to define things that can occur regardless of the odds.
Probability: I use this term to define things that can more likely occur and generally I try to attach a percentage of probability (Odds) of it occurring.

This is the interesting part…IF the MOS numbers are tracking towards a ratio of 15 months for the Bavituximab combo vs. 10 months for placebo combo then my predicted number of events for first look-in indicate:
Bavituximab combo = 30 to 40 events
placebo combo = 120 to 130 events

What are the odds (probability)?
28%...A little better than 1 in 4
That implies that there is, in my view a 28% probability that the trial could be stopped at the first look-in.
Not likely…But possible with a 28% probability.

SECOND LOOK-IN:
Bavituximab combo = 70 to 80 events
placebo combo = 150 to 165 events

WELL NOW…
What are the odds (probability)?
47.6%...Nearly a 50-50 proposition…!!!
Slightly less likely then likely by only 2.4%”

Time will tell…And time is near…

I hope that this might answer some of your questions and if not, please feel free to question, challenge, analyze…Peer review is always welcomed…And necessary.

And as I stated in a previous post:

NOTE TO ALL OTHER READERS…

The above should NOT be used as investment advice…In previous posts I have labeled my projections as an exercise in mental_masturbation and written for my own amusement.

Regards
golfho