Replies to post #27408 on Biotech Values

[DewDiligence]

The same formula can be applied to trials with multiple doses, multiple time points, multiple observations of any kind.

[poorgradstudent]

>I've always wondered if there was some rule of thumb to interpret subgroup significance in trials and published research , and the one above is simple enough even for me !<

I thought it was a brief and informative article. The adjustment (0.05 / K) is a worthwhile rule of thumb. But it has a major limitation when applied to clinical trial results from public / private companies: if a trial hits the primary endpoint, then the sponsor has no trouble reporting it and very rarely mentions any subgroup analyses. So you need not make any adjustment. But when the trial misses the primary endpoint and a magically significant subgroup shows up, it is impossible for investors to know how many analyses were done to find it... thereby leaving K as undefined. I guess we could safely consider K being > 2 for most trials were a subgroup analysis is touted, but my gut feeling says that K is rarely less than 10 in these situations.

This is probably why I run screaming in the other direction when secondary / subgroup analyses are offered as primary evidence of clinical efficacy ;-)

===

I thought the best line from the article was: "When an interaction test for a baseline variable fails to reach the appropriate threshold for significance, conclusions about a differential treatment benefit related to this variable should be avoided or presented with caution."