Bioequivalence and Bioavailability Forum • drugs of long half life

Ken Peh
★

Malaysia,
2013-03-31 08:43
(4402 d 10:13 ago)

Posting: # 10316
Views: 9,164

drugs of long half life [Design Issues]

Dear All,

I have learned from this forum (previous postings) that parallel design is recommended for drug of long half life. I have a request from sponsor to carry out BE on flunarizine. The comparator product (Sibelium, JANSSEN-CILAG S.P.A., Italy) reported in their product insert an elimination half-life of 18 days. To fulfill the washout period of 5 half-life, the washout period will be about 90 days. Does it make sense to run a 2-way cross-over BE study with washout period of 90 days :confused:

? No problem with sampling time as there is guideline on the use of truncated AUC0-72hr.

If parallel design is used, intrasubject CV, period effect, etc can not be estimated.

Your comments are highly appreciated especially those who have experience with this drug.

Thank you.

Regards,
Ken

jag009
★★★

NJ,
2013-04-02 17:17
(4400 d 01:39 ago)

@ Ken Peh
Posting: # 10327
Views: 7,784

drugs of long half life

Post reply

Hi Ken,

I once ran 2 2-way crossover studies with a 90 day washout (management didn't want to take a chance with a parallel study design). The studies were okay, we had 5-10 dropouts.

The sample size for parallel design (I think Helmut mentioned this to me before, right Helmut? :-)

) is based on the total CV (intersubject CV and intrasubject CV).

90% CI computation will be done using the intersubject CV¹

------

1. Niazi, SK. Handbook of Bioequivalence Testing. Drugs and Pharmaceutical Sciences, Vol 171, p28.

Helmut
★★★

Vienna, Austria,
2013-04-02 17:34
(4400 d 01:22 ago)

@ jag009
Posting: # 10328
Views: 7,840

drugs of long half life

Post reply

Hi John!

❝ The sample size for parallel design (I think Helmut mentioned this to me before, right Helmut? :-) ) is based on the total CV (intersubject CV and intrasubject CV).

Yessir.

❝ 90% CI computation will be done using the intersubject CV

Nope. The residual variance in a parallel design is also the total. What Niazi writes:

The width of the conﬁdence interval is determined by the within-subject variance (between-subject variance for parallel group studies) and the number of subjects in the study.

is wrong (or another example of sloppy terminology). In a parallel design between-subject variance is not accessible – only total (or pooled if you prefer).

@Ken:

❝ The comparator product […] reported in their product insert an elimination half-life of 18 days. To fulfill the washout period of 5 half-life, the washout period will be about 90 days.

Don’t fall into the trap of basing you design on a reported mean value. Take the variability into account (i.e., conservatively assume a longer half life). Does the SmPC contain data on the variability (±SD)?

—
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

jag009
★★★

NJ,
2013-04-02 18:01
(4400 d 00:55 ago)

@ Helmut
Posting: # 10329
Views: 7,782

drugs of long half life

Post reply

Hi Helmut,

❝ ❝ 90% CI computation will be done using the intersubject CV

❝

❝ Nope. The residual variance in a parallel design is also the total. What Niazi writes:

The width of the conﬁdence interval is determined by the within-subject variance (between-subject variance for parallel group studies) and the number of subjects in the study.

is wrong (or another example of sloppy terminology). In a parallel design between-subject variance is not accessible – only total (or pooled if you prefer).

Agreed

There is only one value for us to use anyway.
Question, the anova model (GLM) will only contain 1 factor, correct?

Thanks
John

Helmut ★★★ Vienna, Austria, 2013-04-02 18:40 (4400 d 00:15 ago) @ jag009 Posting: # 10330 Views: 7,789	drugs of long half life Post reply
	Hi John, ❝ Question, the anova model (GLM) will only contain 1 factor, correct? Yep – I would suggest `treatment`. — 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

d_labes
★★★

Berlin, Germany,
2013-04-04 10:33
(4398 d 08:22 ago)

@ Helmut
Posting: # 10340
Views: 8,045

parallel FDA

Post reply

Dear Helmut, dear John,

❝ ❝ Question, the anova model (GLM) will only contain 1 factor, correct?

❝

❝ Yep – I would suggest treatment. :-D

A one-way ANOVA (GLM) evaluation will not fit the FDA guideline (see this post long ago).
It is equivalent to the assumption of equal variances.

If you only have 2 groups you can revert to the Welch t-test.
Or you must code something with Proc MIXED (if you speak SASenese :cool:

)

—
Regards,

Detlew

jag009 ★★★ NJ, 2013-04-05 20:01 (4396 d 22:54 ago) @ d_labes Posting: # 10360 Views: 7,795	parallel FDA Post reply
	Hi Detlew, ❝ If you only have 2 groups you can revert to the Welch t-test. ❝ Or you must code something with Proc MIXED (if you speak SASenese ) Thanks. I remember looking at a parallel BE study report and the stat was T-Test. Weekend assignment for me.. PROC MIXED heheh. John

jag009
★★★

NJ,
2013-04-05 23:10
(4396 d 19:45 ago)

@ d_labes
Posting: # 10363
Views: 7,763

parallel FDA

Post reply

Hi Detlew,

❝ Dear Helmut, dear John,

❝

❝ ❝ ❝ Question, the anova model (GLM) will only contain 1 factor, correct?

❝ ❝

❝ ❝ Yep – I would suggest treatment. :-D

❝

❝ A one-way ANOVA (GLM) evaluation will not fit the FDA guideline (see this post long ago).

❝ It is equivalent to the assumption of equal variances.

❝

❝ If you only have 2 groups you can revert to the Welch t-test.

❝ Or you must code something with Proc MIXED (if you speak SASenese :cool: )

Something like this? :-D

Proc mixed data=s1;
class subject trt;
model lcmax=trt / ddfm=satterth;
lsmeans trt /pdiff cl alpha=.1;
estimate 'T vs R' trt 1 -1/ cl alpha=0.1;

John

d_labes
★★★

Berlin, Germany,
2013-04-08 10:29
(4394 d 08:27 ago)

@ jag009
Posting: # 10366
Views: 7,604

random repeated parallel FDA

Post reply

Dear John,

❝ Proc mixed data=s1;

❝ class subject trt;

❝ model lcmax=trt / ddfm=satterth;

❝ lsmeans trt /pdiff cl alpha=.1;

❝ estimate 'T vs R' trt 1 -1/ cl alpha=0.1;

Seems you are on the right path :cool:

.
But I have the feeling that the ddfm option is only half of the truth. Your code gives still only one variance term. Try it.

I must confess that I never had worked out a mixed model to full detail which covers the 2-group parallel as special case. No need up to now.
But I think there had to be somefink like a random or repeated statement to specify different variabilities for test and reference.

The following code seems to work (gives 2 variance parameters) but has to be verified / validated:

Proc mixed data=s1;

  class subject trt;

  model lcmax=trt / ddfm=satterth;

  repeated trt / group=trt;

  estimate 'T vs R' trt 1 -1/ cl alpha=0.1;

run;

—
Regards,

Detlew

jag009
★★★

NJ,
2013-04-08 17:36
(4394 d 01:20 ago)

@ d_labes
Posting: # 10370
Views: 7,760

Proc Mixed: random repeated parallel FDA

Post reply

Hi Detlew,

❝ The following code seems to work (gives 2 variance parameters) but has to be verified / validated:

❝ Proc mixed data=s1;


❝   class subject trt;

❝   model lcmax=trt / ddfm=satterth;

❝   repeated trt / group=trt;

❝   estimate 'T vs R' trt 1 -1/ cl alpha=0.1;

❝ run;

Enlight me guru. Why the need for a repeated trt/group=trt statement though?
We are looking at a parallel study design whereby two groups of subjects are given either T or R in the 1 study period.

Personally I don't see what additional variability(ies) one can extract from a parallel study other than inter-subject CV since we don't give the drugs to the same person twice. The use of total variability (intra and inter) to compute the sample size I guess is just to cover all ground and increase the level of comfort (My old friend always told be liberal on sample size estimation, but then be rational)... Maybe I am wrong and still need to understand more...

Thanks
John

d_labes
★★★

Berlin, Germany,
2013-04-08 18:41
(4394 d 00:14 ago)

@ jag009
Posting: # 10371
Views: 7,838

Proc Mixed: parallel groups with different variabilities

Post reply

Dear John,

❝ We are looking at a parallel study design whereby two groups of subjects are given either T or R in the 1 study period.

Correct.

❝ Personally I don't see what additional variability(ies) one can extract from a parallel study other than inter-subject CV ...

If this should be read total CV, correct.
But the two groups of subjects (under Test or Reference treatment) would allow the separate estimation of CV(tot.)_T and CV(tot.)_R (or s2_T, s2_R in the log domain). And this should be reflected in the overall model.

A Proc GLM ANOVA or your Proc MIXED code assumes equal variabilities aka homogeneous variances, analogous to a t-test assuming equal variances.

But again: My code is a quick shot and may be wrong (or as the Admin of this forum always touts "... should be taken with a grain of salt").

—
Regards,

Detlew

jag009
★★★

NJ,
2013-04-08 22:22
(4393 d 20:34 ago)

@ d_labes
Posting: # 10374
Views: 7,666

Proc Mixed: parallel groups with different variabilities

Post reply

Hi Detlew,

❝ ❝ Personally I don't see what additional variability(ies) one can extract from a parallel study other than inter-subject CV ...

❝

❝ If this should be read total CV, correct.

❝ But the two groups of subjects (under Test or Reference treatment) would allow the separate estimation of CV(tot.)_T and CV(tot.)_R (or s2_T, s2_R in the log domain). And this should be reflected in the overall model.

I see what you mean :-)

Yes there should be a intersubject CV for T and R separately since we have two pools of subjects given either T or R.

Thanks
John

Ken Peh
★

Malaysia,
2013-04-04 17:03
(4398 d 01:52 ago)

@ jag009
Posting: # 10343
Views: 7,684

drugs of long half life

Post reply

Dear John,

Thank you for your advice and guidance.

I have searched for the text book suggested by you.
(Niazi, SK. Handbook of Bioequivalence Testing. Drugs and Pharmaceutical Sciences, Vol 171, p28).

However, I do not find much information on parallel study design. I would like to learn more about parallel study, the statistics (what type of statistical model used eg. ANOVA :confused:

), calculations and interpretations.

Regards,
Ken

jag009
★★★

NJ,
2013-04-05 19:45
(4396 d 23:11 ago)

@ Ken Peh
Posting: # 10359
Views: 7,628

drugs of long half life

Post reply

Hi Ken,

Specifically what are you looking for? The sections in that book seem sufficient in terms of study design and the approach. I agree that the stat description is a little vague as it only highlights variability, sample size and what is needed to calculate the 90%CI (Which I agree with Helmut that the wording is wrong).

Are you using WinNonlin or SAS? I recall WinNonlin can perform parallel design analysis (See D_labes response in this thread about SAS). I think there have been several discussions about parallel design in Helmut's crib here ( :cool:

) so you might want to search around.

John
Thanks

John