Bioequivalence and Bioavailability Forum • can inter sub CV%

stat_be
●

2007-09-27 15:15
(6423 d 21:37 ago)

Posting: # 1138
Views: 13,470

can inter sub CV% > intra sub CV%? [Power / Sample Size]

Dear members,

As far as bioequivalence trial is concerned, I wanted to clarify my doubts related to inter subject CV% and intra subject CV%.

Can inter subject CV% greater than intra subject CV? if yes then in what situation it may happen and what conclusion can be drawn on that basis?
what are the formulas to calculate to both of them from GLM Anova?

Seniors, please clear my doubts as I am new to this field.

thanks in advance

stat_be

Helmut
★★★

Vienna, Austria,
2007-09-27 15:34
(6423 d 21:18 ago)

@ stat_be
Posting: # 1139
Views: 11,132

CVinter > CVintra!

Post reply

Dear stat_be!

❝ […] my doubts related to inter subject CV% and intra subject CV%.

So you are talking about a cross-over design.

❝ 1. Can inter subject CV% greater than intra subject CV? if yes then in what situation it may happen and what conclusion can be drawn on that basis?

Yes, this is the rule. In a cross-over design the PK responses within a particular subject are compared (between occasions).
I hardly can imagine a situation where variability between subjects (i.e., Brown vs Blair) is lower than within a subject (i.e., Brown on day 1 vs Brown on day 2). Although theoretical possible from a statistical point of view, such a behaviour is counterintuitive, and IMHO not a single case is described in the literature.
BTW, the expectation of CV_intra < CV_inter leads to the general preference of cross-over designs over parallel designs in BE studies.

❝ 2. what are the formulas to calculate to both of them from GLM Anova?

For formulas see this post.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

stat_be
●

2007-10-03 09:35
(6418 d 03:17 ago)

@ Helmut
Posting: # 1151
Views: 10,975

CVinter > CVintra!

Post reply

Dear HS, thanks

❝ So you are talking about a cross-over design.

Yes. It is a cross-over design for pilot be of 12 subjects.

❝ Yes, this is the rule. In a cross-over design the PK responses within a particular subject are compared (between occasions).

❝ I hardly can imaginge a situation where variability between subjects (i.e., Brown vs Blair) is lower than within a subjecti.e., Brown on day 1 vs Brown on day 2). Although theoretical possible from a statistical point of view, such a behaviour is counterintuitive, and IMHO not a single case is described in the literature.

❝ BTW, the expectation of CV_intra < CV_inter leads to the general preference of cross-over designs over parallel designs in BE studies.

Please find below data for Tmax of a drug XXX and suggest. I have used proc glm for this and here MSE_Sub(seq) is less than MSE_residual so inter subject CV% can not be calculated. This is practical data obtained. Please let me know that which kind of conclusion should be made on the basis of this data. Same thing happened in the data of Cmax where %CV_inter < %CV_intra.

Sub Seq   Tmax_P-I  Tmax_P-II

 1   BA      3.50      2.50

 2   BA      1.50      3.00

 3   AB      2.50      5.00

 4   AB      3.50      2.50

 5   BA      5.00      2.50

 6   AB      2.50      2.50

 7   AB      2.50      2.50

 8   BA      3.00      4.50

 9   BA      4.00      4.00

10   AB      1.50      2.00

11   BA      2.50      3.00

12   AB      3.00      3.00

Please discuss as so many of new comers in the feild might be facing this kind of problem.

Thanks in advance.

Stat_BE

--
Edit: Tabs removed from table; original quotes restored (see here). [HS]

Ohlbe ★★★ France, 2007-10-03 12:04 (6418 d 00:48 ago) @ stat_be Posting: # 1152 Views: 11,023	CVinter > CVintra! Post reply
	Dear stat_be, ❝ Please find below data for Tmax of a drug XXX and suggest. ANOVA for Tmax ??? ❝ Same thing happened in the data of Cmax where %CV_inter < %CV_intra. Could you rather provide these ? Looks more relevant to me... Regards Ohlbe

Helmut
★★★

Vienna, Austria,
2007-10-03 17:02
(6417 d 19:50 ago)

@ Ohlbe
Posting: # 1158
Views: 11,445

ANOVA not for Tmax!

Post reply

Dear stat_be!

❝ ❝ Please find below data for Tmax of a drug XXX and suggest.

❝ ANOVA for Tmax ??? :surprised:

Ohlbe is right; t_max must not be subjected to any statistical method (e.g., GLM, ANOVA, t-test, F-test,…) requiring continuous variables.
You have to apply a nonparametric method (additive model / untransformed data)* instead - which results in a PE ±0.00, 90% CI [-1.00, +0.75] from your data.

Hauschke D, Steinijans VW and E Diletti
A distribution-free procedure for the statistical analysis of bioequivalence studies
Int J Clin Pharm Ther Toxicol 28(2), 72–8 (1990)

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

Helmut
★★★

Vienna, Austria,
2007-10-15 18:15
(6405 d 18:37 ago)

@ stat_be
Posting: # 1190
Views: 11,527

CVinter > CVintra!

Post reply

Dear stat_be!

❝ […] I have used proc glm for this and here MSE_Sub(seq) is less than MSE_residual so inter subject CV% can not be calculated. This is practical data obtained. […] Same thing happened in the data of Cmax where %CV_inter < %CV_intra.

It would be nice, if you could provide us with you C_max data!

Just a hint from Pharsight’s WinNonlin knowledge base:
Solution for WinNonlin Bioequivalence Warning 11094: Negative final variance component
The negative final Variance Component warning most likely indicates that, if using Subj(Seq) as a random effect, the within-subject variance (residual) is greater than the between-subject variance. Probably a more appropriate model is to move Subj(Seq) out of the random model and into the fixed model, i.e.,

Sequence+Subject(Sequence)+Formulation+Period.

To do this, use the following steps:

Start BE Wizard.
Specify reference formulation → Next
Specify dependent variable, e.g. Cmax → Next
Click Delete Random to remove Subject(Sequence) component → Back
Add Subject(Sequence) to the Fixed effects model as shown above
Click Calculate

You can reformulate you model in SAS accordingly.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

stat_be ● 2007-10-16 14:19 (6404 d 22:33 ago) @ Helmut Posting: # 1199 Views: 11,093	CVinter > CVintra! Post reply
	Dear HS, Thanks for ur advise. Please let me know the adjustment in proc mixed insted of proc glm????? thanks in advance.

Helmut ★★★ Vienna, Austria, 2007-10-16 15:15 (6404 d 21:37 ago) @ stat_be Posting: # 1200 Views: 11,093	CVinter > CVintra! Post reply
	Dear stat_be! ❝ Please let me know the adjustmnet in proc mixed insted of proc glm????? Sorry to see so many questions marks, but I'm not gifted with — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes

krishna
☆

India,
2012-11-30 06:46
(4533 d 05:06 ago)

@ Helmut
Posting: # 9639
Views: 8,771

CVinter > CVintra!

Post reply

Hi Helmut,

❝ Sequence+Subject(Sequence)+Formulation+Period.

❝

❝ To do this, use the following steps:

❝ 1. Start BE Wizard.

❝ 2. Specify reference formulation → Next

❝ 3. Specify dependent variable, e.g. Cmax → Next

❝ 4. Click Delete Random to remove Subject(Sequence) component → Back

❝ 5. Add Subject(Sequence) to the Fixed effects model as shown above

❝ 6. Click Calculate

After doing above correction in WinNonlin, I am observing slight increase in SE as a consequence CI is widening slightly. Also, p-value for sequence is working against 'Residual Error' as error term rater than 'Subject(sequence)' term. How this problem can be fixed and what is the reason for increment of SE ?

Thanks,

Krishna.

Helmut
★★★

Vienna, Austria,
2012-11-30 14:54
(4532 d 20:58 ago)

@ krishna
Posting: # 9643
Views: 9,024

WinNonlin random vs. fixed model

Post reply

Hi Krishna!

❝ After doing above correction in WinNonlin, I am observing slight increase in SE as a consequence CI is widening slightly.

Which version of WinNonlin? Did you get a “Warning 11094: Negative final variance component” in the random effects model (WinNonlin’s default)? If yes, the result is not reliable (overspecified model). Therefore results might differ – but only the fixed effects model is valid. If you didn’t get a warning and use the fixed effects model (according to EMA’s GL) the results should be identical.

❝ Also, p-value for sequence is working against 'Residual Error' as error term rater than 'Subject(sequence)' term.

I don’t know why you want to test for this effect at all. BTW, you are rather interested in Subject(Sequence), which is identical in both models. Try it with WinNonlin’s Data22 example (the famous Clayton & Leslie data set).*

❝ How this problem can be fixed and what is the reason for increment of SE ?

Don’t understand what you want to ‘fix’. Consider registering at Pharsight’s Extranet and ask there.

Clayton D and A Leslie
The bioavailability of erythromycin sterate versus enteric-coated erythromycin base taken immediately before and after food
Int Med Res 9, 470–7 (1981)

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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