Avid Bioservices Inc (CDMO)

[Protector]

Statistical Relevance and the cowboy stories.

FTM already made it clear that the pancreatic CT wasn't setup to show and significance . so the discussion here is more to give an indication why, besides the fact that no significance was sought, PPHM would not give p-value with this top level data and probably will provide them with the detailed data.

Mostly in statistics you will receive the popular 99% correct explanations and the poison sits in the 1% that is omitted.

So let me try to explain it and related it to PPHM pancreatic clinical trial.

When a clinical trial is executed is becomes by definition subject to what is called "random error". This is nothing that is proper to a clinical trial but is true for many statistical applications.

Historically one of the first and main criteria, by agencies such as the FDA in the US, to allow a drug on the market is that it must have 'working substance' and that good results must be due to that working substance under investigation. Therefor they must eliminate the factor of chance and in the case of clinical trials also the so called 'placebo effect'.

The placebo effect can be eliminated by a SOC+placebo control arm, as we had in our 2nd ln NSCLC or by any formula where the placebo effect is canceled out in all arms.
The chance factor is eliminated by assigning a p-value that tells us about statistical relevance, in our case statistical relevance of the clinical trials results.

A result of a clinical trial is statistically significant if it is so good that such result is considered to occur only on rare occasion by chance, in other words something cancels out randomness and that something is the drug under investigation, resulting in the results showing up consistently otherwise then it would have it is was obtained by pure chance.

To put it in a more non clinical trial example, if you win the lottery 5 times in a row, while it is already extremely rare that you would win it once, then that is NOT statistically relevant and it is very likely that you are cheating in some way or another because winning it 5 times in a row by chance has an extremely small probability.

So far so good. But there is another rule that applies to the theory that says that external factors may not impact the results. And in the case of a clinical trial that isn't the case.

The patient assignment to the arms have a huge impact on the results as was demonstrated a few posts ago. Would the patients FIRST have been grouped in ECOG groups and each arm would receive an equal number of 0,1 and 2 patients then the external factor would have been eliminated. Hard to do because it would imply that you know the ECOG of all patients that are going to show up during the enrollment period in advance, which isn't the case. Assigning them on the fly with a toggling technique is possible but would open the clinical trial to a certain degree of 'fixing' because in a non-blinded trial if two patients of a same ECOG group show up one could systematically assign the less sick one to the Bavi arm and the other to the control arm. To make this possible an provision that isn't available today for clinical trial would be needed.

To compare it with the lottery, assume there are numbers that always win, some never and some in 50% of the time instead of all numbers having an equal chance. If the Lottery Company would pull each time from an equal amount of numbers of each category everything would still be fine.
But if they randomly assign a collection of numbers to the draw of each category and that changes each week then there are going to be weeks where it is very hard to win and week where it will be very easy to win. In other words the random assignment of numbers equals the assignment of patients to the clinical trial arms.

As a consequence PPHM has all interest to provide p-value at a moment where they can FULLY attract the attention to the imbalance of ECOG=2 patients in the pancreatic trial. There where 3 times more ECOG 2 patients in the Bavi arm then in the Control arm. Hence, PPHM entered a lottery with lots of numbers that could not win and they where mainly in the Bavi arm.

Remember, statistical relevance isn't really an issue in this clinical trial because we are not seeking to show relevance and this post was merely to explain why it was prudent from PPHM not to throw the p-values in an hostile market with eMagazines that wouldn't hesitated to present these p-values without any contextual clarification and so painting a wrong impression.

Now they couldn't get much further then to say the p-values must be poor because PPHM didn't provide them. Every one will at least recognize that as speculation.