Replies to post #5088 on Dendreon Corp fka DNDNQ

[walldiver]

I don't know if that would be allowable, and I haven't heard of it being done before. Not to say it hasn't been done before, of course.

[DewDiligence]

>Wouldn't it be wiser then for DNDN to set the interim look based on two factors being achieved: 180 deaths AND all patients being enrolled for a set minimum time period? I realize this would push the date out for the interim look significantly, but it would also give the look better odds of being stat sig.<

From a powering standpoint, such a dual trigger for the interim look would be tantamount to using the ordinary number-of-events trigger with a larger number of events.

Of course, the likelihood of hitting a given p-value threshold would be higher with more events but, as you pointed out, it would take longer for the trigger to be triggered.

Deciding when to conduct an interim look is a tradeoff between timing and statistical power. If an interim look is conducted so late that the final look is not much later, the benefit of the interim look from a business standpoint is relatively small even when the interim analysis is successful.