Replies to post #32079 on Biotech Values

[poorgradstudent]

gtop:

>To see this another way, pretend you are the executive in charge of designing the pivotal-trial program for company X. The FDA offers you two options for an SPA that assures approval of your drug if it meets the requirements set forth in the SPA; however, you must pre-select which of the two options to incorporate into the SPA:

1) Success in two identical trials with N patients each and a p-value threshold in each trial of 0.05.

2) Success in a single trial with N patients identical in design to the trials in option #1 except that the p-value threshold is 0.01.<

Where's the third and easiest choice? Run a survival trial and unblind at a single point like most other companies do, all the while keeping your p threshold at 0.05.

[gofishmarko]

>>> 1) Success in two identical trials with N patients each and a p-value threshold in each trial of 0.05.

2) Success in a single trial with N patients identical in design to the trials in option #1 except that the p-value threshold is 0.01.

--
Which option would a reasonable person select?

Option #2, of course. <<<

I have to admit I wasn't thinking about this aspect of the GTOP discussion , and since the single trial has a greater chance of a Type 1 error than the combo of two trials , it makes sense that it has a reduced chance of a Type 2 error , but it's not intuitively obvious to me that N could remain the same. If the two single trials each had a power of .9 with N patients each and the combined power ( for the needed two successes ) was .9x.9=.81 , does it follow that the single trial , with the same N , would have a power > .81 ?

I could definitely use a statistics refresher course. The problem is that there's a list of refresher courses I need , and the list is growing daily.