Replies to post #95124 on Biotech Values

[ocyan]

Clark - I am disappointed that you, as a purported physicist, would use the term "scientific method" with respect to Dew's way of doing things.

When "discount" which implies a computation is mentioned, I expect a well-defined procedure justified by a properly expressed theory to carry about the computation. You don't just say "I have this magical PSB thing that tells me that the HR is going to be xxx". Didn't even you yourself disagree with Dew's HR 1.20 and later 1.25?

It should also be said that all of those predictions were done without Dew actually knowing what the triggers for the interim and final looks of IMPACT were. That meant that he likely assumed constant HR, an assumption that you know to be wrong.

The final-final logrank HR is now known to be close to 1.33 which improved from the 1.29 seen at the final look. If the trial continued longer, it is feasible that the HR will come close to 1.45 than the pathetic 1.25 that Dew predicted.

Somewhere in our discussions on the IV board about PSB, I have said that it is ok as a rule of thumb when you have no data to model a trial with. But in the case of IMPACT, data were plentiful, simply use a rule of thumb instead of doing some calculations would be wrong.

I maintain that until Dew or you outline a theory and give me a procedure with well-defined parameters that I could use to discount a chance of a trial, PSB must be treated in the same way that a Tarot card reader makes predictions.

In the biotech space, predicting a failure always has a higher probability of success than predicting a success. A good Tarot reader would know that too and adjust her predictions accordingly.

This is my absolute last post here.

Best Regards.

[DewDiligence]

DNDN GNVC:

…Dew did adjust [the forecasted HR] from 1.20 at some point to 1.25 - but I don't remember when relative to the interim.

My 1.25 forecast was in Apr 2008, many months before the interim analysis was made public (#msg-28441649).

For Ocyan to come over here to try to spin his bloated HR forecast of 1.45+ as a stroke of genius shows how far he will obfuscate when his mathematical models turn out wide of the mark (#msg-49710195).

p.s. With DNDN, Ocyan was at least in the right ballpark; with GNVC’s PACT study of TNFerade, Ocyan’s model was not even in the right zip code.

As long as Ocyan continues to deny that program-survival bias is a real phenomenon—and his post on iHub today gives no indication that this will change—Ocyan’s models will continue to be biased high relative the true efficacy of the therapies being modeled. The inevitable outcome is that Ocyan’s models will produce relatively few good forecasts and relatively many boneheaded ones, such as the one he made for GNVC.