Biotech Values

[DewDiligence]

DNDN – Let’s run some pseudo HR numbers:

>[PGS to iwfal]: If they need a 22% reduction in the risk of death at final analysis, and you believe the curves get stronger from here on out, it seems to be a good bet, no?<

Let’s define pseudo HR as a first-order linear approximation for the HR during a given portion of a trial (in this case 9902b).

Without loss of generality, we may assume that the randomization ratio between the trial arms is 1:1. (This simplifies the arithmetic, but doesn’t alter the calculated pseudo-HR — try it and you’ll see!)

If we assume for the sake of discussion that the interim event trigger in 9902b was 240 deaths (a number that’s been used by various DNDN posters from time to time), the known HR=1.25 at the time of the interim analysis gives to a first-order linear approximation that the 240 deaths are comprised of 133 deaths in the control arm and 107 deaths in the Provenge arm. (133/107=1.243, which is the closest match to 1.25 using whole numbers of patients.)

Dr. Gold said on yesterday’s CC that a “22% treatment benefit” is needed at the final analysis for the 9902b trial to be deemed a success under the SPA; hence the required HR at the final analysis is 1/0.78=1.282.

We know that the number of deaths at the final analysis is 304, an increment of 64 from our assumed 240 deaths at the interim analysis. To achieve HR=1.282 at the final analysis, a first-order linear approximation requires that at least 171 of the 304 deaths be in the control arm and at most 133 of the 304 deaths be in the Provenge arm. (171/133=1.2835, which is the closest match to 1.282 using whole numbers of patients.)

171-133=38; hence, a first-order linear approximation requires that at least 38 of the 64 incremental deaths be in the control arm and at most 26 of the 64 incremental deaths be on the Provenge arm.

If we convert these 38 and 26 numbers into a pseudo-HR figure for the period between the interim and final analyses, the required pseudo HR in this portion of the 9902b trial for the trial to be a success is 38/26=1.462.

If the interim event trigger was less than 240, then the actual situation is less dire than the numbers shown above. (The algorithm for calculating the pseudo-HR would remain the same but the plug-in values would be different.)

In any event, the main point is this: Under most sets of reasonable assumptions, hitting the required HR threshold at the final analysis is no cakewalk.