Bioequivalence and Bioavailability Forum • Intrasubject CV in Partial Replicate Design (TRT)

luvblooms4u
●

2010-11-18 07:12
(5698 d 08:50 ago)

Posting: # 6161
Views: 6,956

Intrasubject CV in Partial Replicate Design (TRT) [Design Issues]

Dear All

Greetings :flower:

Few days back, we did a 3 way pilot biostudy (T1 vs R vs T2) of a BCS class 2 molecule.
The results were like this
For Test 1

[image]

For Test 2

[image]

The performance of the product was as per the expectations.
But what hold my attention was % Intrasubject CV, which was same incase of both the formulations though intersubject variation differed.

After looking at this I am a bit confuesd that
1. Whether the ICV calculations for Test1 and Test2 are right or somewhere something I did wrong?
2. Why are we getting the similar ICV where as intersubject variation differs?

P.S: I am not good in stats and question like this make my head spin :lookaround:

Any suggestions or ideas?

Thanks in advance!

Luv

d_labes
★★★

Berlin, Germany,
2010-11-18 15:23
(5698 d 00:39 ago)

@ luvblooms4u
Posting: # 6166
Views: 5,600

X-over for more than 2 treatments

Post reply

Dear Luv,

This study had not a "partial replicate design" but a cross-over design with more than two formulations. See Chapter 10 of
Shein-Chung Chow, Jen-pei Liu
Design and Analysis of Bioavailability and Bioequivalence Studies
3rd ed., Chapman & Hall/Crc Biostatistics Series
The usual evaluation of such a study employs the same ANOVA model as for the classical 2x2 crossover, but now with more than 2 treatments, more than 2 periods and (usually) more than 2 sequences. This will give you one MSE as an estimate of the intra-subject variability and therefore one CV as you noticed by yourself.
If you follow the recommendation of the new EMA guidance (see page 14 under Subject accountability) you have to drop the data of the formulation not interested in for each comaparision (i.e. use only the data for T1 and R in comparing T1 vs. R, for instance). Then you will get different CV's for each comparision.
But IMHO this is statistical not correct, not to say nonsense .
Moreover it violates the usual regulatory attitude that only in exceptionally cases data could be excluded from analysis.
To have estimates of the intra-subject variabilities specific for the formulations one has to resort to a study with a replicate design.
The partial replicate design is a design with 2 treatments (T and R), 3 periods and the sequences TRR/RTR/RRT. As you can see, in this design only the reference is administered more than once.
What you call inter-subject CV is simply the CV calculated for the values of the PK metrics from each formulation alone I suppose. This is total CV.

Hope this helps.

—
Regards,

Detlew

jag009
★★★

NJ,
2014-09-25 18:03
(4290 d 22:59 ago)

@ d_labes
Posting: # 13586
Views: 4,102

Revisit: X-over for more than 2 treatments

Post reply

Hi D_labe,

❝ • If you follow the recommendation of the new EMA guidance (see page 14 under Subject accountability) you have to drop the data of the formulation not interested in for each comaparision (i.e. use only the data for T1 and R in comparing T1 vs. R, for instance). Then you will get different CV's for each comparision.

I got some free time to goof around here... Question, what statement(s) do you use in SAS (Proc GLM) to do the above, i.e, drop the data of the formulation not interested in for each comparison? Or you meant removing the data from that treatment completely from the dataset?

Thanks
John

d_labes
★★★

Berlin, Germany,
2014-09-26 16:26
(4290 d 00:36 ago)

@ jag009
Posting: # 13595
Views: 4,039

Dropping data in analyzing X-over for more then 2 treatments

Post reply

Hi John,

❝ ... Question, what statement(s) do you use in SAS (Proc GLM) to do the above, i.e, drop the data of the formulation not interested in for each comparison? Or you meant removing the data from that treatment completely from the dataset?

as always there are many ways to Rome:

A where statement in Proc GLM retaining only the data necessary for the pairwise comparision
A datastep before the call of Proc GLM dropping the data not involved in the pairwise comparision under consideration

My SAS programs / macros use the latter approach. But this is only a matter of taste.

—
Regards,

Detlew