Replies to post #65956 on Biotech Values

[DewDiligence]

>BIOD – It's not quite fair to say that it was only stat sig by excluding India, as that was the Type I trial only (and type II is obviously the "market").<

Agreed.

>OTOH, I suspect the real issue is that it was stat sig *inferior* to Humulin, despite being stat sig non-inferior (yes, I think I said that right).<

Which of the two trials are you referring to now?

>How common is that [for a trial to be statsig non-inferior as well as statsig inferior]?<

This seemingly paradoxical outcome is entirely possible from a mathematical standpoint. However, few non-inferiority trials contain a prespecified two-sided test for superiority/inferiority of the experimental agent vs the control; without such a prespecified test, one cannot legitimately declare the experimental agent inferior to the control even if it comes out that way according to a post hoc analysis.

>It's probably not a good idea to have an also ran in this market.<

Agreed. The various insulin analogs (what Novo-Nordisk calls “modern insulins”) are steadily gaining market share at the expense of ordinary insulin. A tweaked formulation of ordinary insulin, even if it works, is not exactly a winning business idea, IMO. Also see #msg-31043957 for related commentary.

[DewDiligence]

Addendum re BIOD and non-inferiority analyses:

>How common is that [for a trial to be statsig non-inferior as well as statsig inferior]?<

Following up on the comments in #msg-32024599: There’s a notable case where the combination of statsig non-inferior and statsig inferior is clearly impossible:

1. The finding of non-inferiority requires that the good-side bound on the 95% CI for the treatment difference be on the good side of zero. (This is sometimes, but not always, a requirement specified in a statistical-analysis plan for non-inferiority.)

2. There is a two-sided test for superiority/inferiority rather than a one-sided test for superiority per se.

3. The non-inferiority test and the superiority/inferiority test are conducted on the same endpoint using the same methodology.

--
For a finding of inferiority, #2 above requires that the good-side bound on the 95% CI for the treatment difference be on the bad side of zero, and hence #2 is incompatible with #1. It follows that the treatment in question can’t be both statsig non-inferior and statsig inferior.