https://www.linkedin.com/pulse/why-have-unequal-randomisation-rat...

First 57 Last

Share
Profile

Follow
truss20

816 Posts.

332

23/03/24

08:55

Post #: 73023066
Share

Originally posted by JB1975: ↑

I was sure I heard Killian say it in an interview, however I can't recall which one.
If he said 30 30 specifically, since becoming CEO specifically, as opposed to saying it when Ross was CEO, then I am wrong. But I don't think I'm wrong.

I disagree, though (), that a split 30:30 or 40:20 matters much statistically. No offense meant here but the handling of statistics in relation to GvHD trials is something I've seen ODAC expert panelists get wrong - Susan Halibi (or Habibi) (not sure of her name spelling but certain she was on the panel for her statistical expertise and her statistical comments suggested MSB didn't have enough of a case) - and yet even she voted Yes. She set her own expertise aside - she couldn't have done her homework - she couldn't have known then (Aug 2020) what the FDA knew even then and even years later (when they rejected BLA 2).

Overwhelmingly the holders on MSB are still holding and hoping against the stats - they aren't able to see that there isn't enough data - that there cannot be.

I've also seen the MEND trial go through with what I considered to be egregious design errors - those never became public.

I know there are supposed to be statisticians involved in checking as things go along, And I know you know that - to some extent - the DSMB I know you think you know should check, but just as there are supposed to be auditors of company records etc - I personally don't trust them. I think they check only specific things they don't check everything. And I don't think there checking requirements are as broad as you think they are. I think they do the work the party paying them to do gets them to do.

I can't give you a really sound reason for my concern here that I'd be sure you could follow - but I am myself quite sure 30 30 and 40 20 will matter. Mathematically it must matter for confidence in a week or marginal result. And even in making a strong result statistically stronger.

The patients chances of being assigned to the control arm are twice as high in one of those scenarios 1 chance in 2, (30 c 30 t) versus 1 chance in 3 (20 c 40 t) and the control arm is corticosteroids (and no other treatments including other test treatments for the duration - or those other treatments would become confounders of the study) - so thats a commitment for patients going in to accept they cannot take another go at another trial (doing so would invalid the stats for this one) should they feel themselves that they are not respond to the treatment (they won't know if they got treatment or control - but if they get poor outcomes they may assume they got the control and are doing no better than if they just got corticosteroids and it isn't working for them and so want to try something else).

A lot of words in this post - I doubt they'll persuade you though.

Expand

https://www.linkedin.com/pulse/why-have-unequal-randomisation-ratios-clinical-trials-joshua-brostoff

Regarding randomisation split:

So 60 patients vs 67. That’s not a huge difference.

CYP Price at posting: 19.0¢ Sentiment: Buy Disclosure: Held
3 Upvote

1 Great analysis

Reply