Replies to post #138561 on Biotech Values

[iwfal]

A strict OBF type boundary would have a nominal p-value of 0.043. You can't just apply the fixed sample size function by tinkering the alpha until you get the 705 event goal, right? Also, the smaller the interim spend, such as a haybittle-peto function, the less sample size inflation, the smaller the event goal.

I realize that the technique I've described is an approximation only. But one that works reasonably well because typically only a miniscule fraction of the power comes from the interims. (triple whammy - smaller alpha, fewer events, the significant majority of stat sig 75% interims would also be stat sig on final)

As for the alpha that goes to the final - I haven't worried about the exact value (which I believe is somewhere between 0.043 and 0.044) because:

a) I think there is some small amount of noise (e.g. to Lan Demets or not and some other nuances)

b) the difference is a few percent at most to any given assigned value