Helmut
★★★
avatar
Homepage
Vienna, Austria,
2008-08-21 16:10
(5698 d 23:38 ago)

Posting: # 2217
Views: 7,336
 

 'Randomisation' for RR/RR design [General Sta­tis­tics]

Dear all,

I was asked asked to design a pilot study to assess the CVintra of an innovator’s product only (the generic is not ready, actually it’s some kind of feasibility issue).
Problems arise from the ‘pseudo randomisation’ – there’s only one product and sequences in a conventional 2×2 cross-over will be fixed a priori like this:
 S/P |  I  II
-----+-------
  1  |  R1 R2
  2  |  R2 R1

Of course the assignment to ‘R1’ is ‘R2’ is arbitrary; any other then the chosen one in the protocol will result in another PE and CVintra
Any ideas/suggestions?

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2008-08-21 16:32
(5698 d 23:17 ago)

@ Helmut
Posting: # 2218
Views: 5,761
 

 'Randomisation' for RR/RR design

❝ Any ideas/suggestions?


Dear HS, could you clarify: What is R1 and R2 - from your description I don't get it.

EM.
Ohlbe
★★★

France,
2008-08-21 17:17
(5698 d 22:31 ago)

@ Helmut
Posting: # 2219
Views: 5,868
 

 'Randomisation' for RR/RR design

Dear Helmut,

If I get the point correctly, your idea is to reproduce the same ANOVA model as what you would have in a conventional 2x2 study, is that right ? I mean, to introduce in the ANOVA the sequence effect (normally TR or RT) which you wouldn't have in your study as you only have RR ?

If I look at slide 63 of your lecture, sequence effect can be due to either a real sequence effect (you wouldn't have one with a single RR sequence), a failure of randomisation (not with no randomisation), a true carryover effect (you know enough about BE trials to plan your trial properly, at least if the sponsor listens to you :-P ), or a formulation by period interaction (which I'm not sure you would be able to detect with your artifical R1R2 / R2R1 randomisation). Adding a "randomisation" in your trial and in your model would indeed introduce a "sequence" effect in your ANOVA, and therefore reduce the MSE somehow (and then reduce the intra-CV), but wouldn't it just be an artefact ? Couldn't you then be under-estimating your intra-CV ?

I'm just thinking aloud, you know much more about these things than I do.

Best regards
Ohlbe
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2008-08-21 19:22
(5698 d 20:26 ago)

@ Ohlbe
Posting: # 2222
Views: 5,806
 

 'Randomisation' for RR/RR design

Dear Ohlbe,

❝ If I get the point correctly, your idea is to reproduce the same ANOVA model as what you would have in a conventional 2x2 study, is that right ?


Yes, but it's bullshit. DLabes already gave a nice explanation in his post.
Thanks for thinking aloud - sometimes you don't to see the wood for the trees... ;-)

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
d_labes
★★★

Berlin, Germany,
2008-08-21 18:59
(5698 d 20:50 ago)

@ Helmut
Posting: # 2220
Views: 5,945
 

 'Randomisation' for RR/RR design

Dear HS,

(emphasis by me)

❝ Problems arise from the ‘pseudo randomisation’ – there’s only one product and sequences in a conventional 2×2 cross-over will be fixed a priori like this:

❝  S/P |  I  II
❝ -----+-------
❝   1  | R1  R2
❝   2  | R2  R1

❝ Of course the assignment to ‘R1’ is ‘R2’ is arbitrary; any other then the chosen one in the protocol will result in another PE and CVintra


I cannot imagine any criterion to decide what R1 and R2 is, if it is the same product, same batch and so on.

What you have is only
  S/P  |  I  II
 ------+---------
   RR  | µR  µR+p

with µR the mean for R (your Reference product) and p a period effect.
Within a subject you have to add an error term to specify the full model.

Thus you cannot and need not randomize any arbitrary sequences.

I would just run an ANOVA with a fixed period effect and a random subject effect. The residual error is then what you are looking for, I think.

Regards,

Detlew
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2008-08-21 19:16
(5698 d 20:32 ago)

@ d_labes
Posting: # 2221
Views: 5,836
 

 'Randomisation' for RR/RR design

Dear DLabes!

❝ I cannot imagine any criterion to decide what R1 and R2 is, if it is the same product, same batch and so on.


Yes, true. I thought I may overcome these problems in ‘fixing’ treatments – of course arbitrarily – in order to stay within a conventional layout.

❝ What you have is only

❝   S/P  |  I  II
❝  ------+---------
❝    RR  | µR  µR+p

❝ with µR the mean for R (your Reference product) and p a period effect.

Within a subject you have to add an error term to specify the full model.


❝ Thus you cannot and need not randomize any arbitrary sequences.


❝ I would just run an ANOVA with a fixed period effect and a random subject effect. The residual error is then what you are looking for, I think.


Thanks, that helped a lot!

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
104 visitors (0 registered, 104 guests [including 11 identified bots]).
Forum time: 14:49 CET (Europe/Vienna)

With four parameters I can fit an elephant,
and with five I can make him wiggle his trunk.    John von Neumann

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5