Bioequivalence and Bioavailability Forum

Amandine
●

2009-06-04 14:31
(5816 d 09:30 ago)

Posting: # 3820
Views: 16,287

4-period BE study fed/fasting: statistical issues [General Statistics]

Dear all,

We are preparing a 4-period single-dose BE study under both fed and fasting conditions (as described in the new draft BE/BA EMEA guidance: Test fast/Ref fast/Test fed/Ref fed).

However, the product we are developing has a great food effect and a high within-subject variability in fed conditions (not under fasting conditions).

Our concern is:

is it possible to compare each product in fed conditions versus fasting conditions (ie, food effect of each formulation) whereas the variances will be greatly different under fed and fasting conditions?
moreover, we think that the residual variability of the 4-way ANOVA will be penalized (strongly increased) by the fed arms. Consequently, we fear not to prove bioequivalence between Test and Reference, even in fasting conditions, because of this high residual CV.

What is your opinion about this design? Is it more risky than two 2-way studies?
Do you confirm the increase of residual variability in such a study and the difficulties to prove bioequivalence under fasting conditions?

Thank you in advance for your answers.

Best regards.

Amandine

Helmut
★★★

Vienna, Austria,
2009-06-04 14:55
(5816 d 09:06 ago)

@ Amandine
Posting: # 3821
Views: 14,850

Confounded effects; common variance

Post reply

Dear Amandine!

❝ We are preparing a 4-period single-dose BE study under both fed and fasting conditions (as described in the new draft BE/BA EMEA guidance: Test fast/Ref fast/Test fed/Ref fed).

You know that a draft is a draft is a draft. ;-)

At the EUFEPS-workshop Kamal Midha and myself strongly argued against the suggested design. Below my comments I have sent to EMEA in January 2009.

It should be noted that in a four-period single dose crossover design study (both products fed and fasted) treatment effects and food effects are massively confounded and no unbiased estimates can be obtained.
In a BE study the main effect of interest is ‘treatment’ (≥2 different formulations, but either in fasting or in fed state). In a food effect study it is ‘food’ (using the same treatment). One of the main assumptions in the usual (nonreplicate) cross-over model is an Independent Identically Distribution (IDD) of effects. This assumption simply may not hold. If e.g., the variability of the reference is higher than the one of the test, one will obtain a high common variance and the test will be penalized for the reference performing badly. For most MR formulations one yet would expect different variabilities in fasting and fed state. Even for IR formulations food will change liver blood flow → hepatic clearance → not only the absorption, but also the elimination may be altered (note: constant clearance is the main assumption in BE). Since the suggested design study is of a nonreplicate design with 2 effects (2 levels: fasting|fed, 2 levels: T|R) the assumption of a common variance is downright absurd.
An alternative to two different studies (T_fasted|R_fasted and T_fed|R_fed), where an inter-study comparison as parallel groups (T_fed vs. T_fasted and R_fed vs. R_fasted) is lacking power, a 2-sequence, 4-period design of following type would avoid confounding issues:
T_fasted R_fasted T_fed R_fed R_fasted T_fasted R_fed T_fed
In such a design treatments in periods 1 and 2 can be compared in fasted state and in periods 3 and 4 in fed state as a conventional cross-over. Additionally T_fed vs. T_fasted and R_fed vs. R_fasted can be evaluated as a paired design (with high power, but avoiding confounding issues).

—
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,
2009-06-10 12:10
(5810 d 11:51 ago)

@ Helmut
Posting: # 3840
Views: 14,654

Neglecting period effects?

Post reply

Dear Helmut,

❝ An alternative to two different studies [...] a 2-sequence, 4-period design of following type would avoid confounding issues:

❝ T_fasted R_fasted T_fed R_fed

❝ R_fasted T_fasted R_fed T_fed

❝ In such a design treatments in periods 1 and 2 can be compared in fasted state and in periods 3 and 4 in fed state as a conventional cross-over.

❝ Additionally T_fed vs. T_fasted and R_fed vs. R_fasted can be evaluated as a paired design (with high power, but avoiding confounding issues).

What exactly You mean with "evaluated as a paired design"?

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2009-06-23 16:38
(5797 d 07:23 ago)

@ d_labes
Posting: # 3894
Views: 14,629

Assumptions...

Post reply

Dear D. Labes,

welcome back from Salzburg!

❝ What exactly You mean with "evaluated as a paired design"?

Well, with the proposed design one avoids confounding issues, but you are right that one must assume lacking period effects!
This may sound weird in the first place, but is regularily used in designs where a single dose phase is followed by a steady state phase in order to assess accumulation: AUC_tau (MD) vs. AUC_inf (SD). In my experience nobody sets up the study in such a way that in ½ of the subjects the study is started with the multiple dose phase first. Any potential period effects in such a design are ignored as well.

—
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

Ohlbe
★★★

France,
2010-03-10 15:27
(5537 d 07:34 ago)

@ Helmut
Posting: # 4889
Views: 13,950

What if...

Post reply

Dear Helmut, dear all,

❝ In such a design treatments in periods 1 and 2 can be compared in fasted state and in periods 3 and 4 in fed state as a conventional cross-over.

❝ Additionally T_fed vs. T_fasted and R_fed vs. R_fasted can be evaluated as a paired design (with high power, but avoiding confounding issues).

Just as a matter of curiosity (and I know there are other difficulties with 4-period trials, such as a higher risk of drop-out). To avoid the problem of assumptions regarding period effects, discussed by D.Labes and Helmut, what would you think of a 4-period trial with the following sequences:


T_fasted  R_fasted  T_fed   R_fed

R_fasted  T_fasted  R_fed   T_fed

T_fed     R_fed     T_fasted R_fasted

R_fed T_fed R_fasted T_fasted

The idea would then be to analyse the data as suggested by Helmut (as two usual 2x2 trials, one for fasted, one for fed) as if there were two groups of subjects in the trial (just like in a large trial when the subjects have to be split due to limited capacities of the facilities), and then to do the "paired design (with high power, but avoiding confounding issues)", without the problem of the sequence effect.

Practically there would certainly be precautions to be taken, such as housing separately the fed and fasting subjects, to avoid confusions (and I wouldn't like to be fasting, and to smell the eggs and bacon of the other guys :-D

), but would it make any sense on a statistical point of view, or is my idea complete nonsense ?

Regards
Ohlbe

—
Regards
Ohlbe

Dr_Dan ★★ Germany, 2010-03-11 16:50 (5536 d 06:11 ago) @ Ohlbe Posting: # 4897 Views: 13,901	What if... Post reply
	Dear all If the drug has a great food effect and a high within-subject variability in fed conditions but not under fasting conditions I would suggest to do two studies: Fasting: two way cross-over study Fed: replicate study Advantage: For the fasted study you will need less subjects and you will have less drop-outs Kind regards Dan — Kind regards and have a nice day Dr_Dan

Helmut
★★★

Vienna, Austria,
2010-03-11 17:39
(5536 d 05:22 ago)

@ Dr_Dan
Posting: # 4898
Views: 14,021

What if…

Post reply

Dear Dan!

❝ […] I would suggest to do two studies:

❝ Fasting: two way cross-over study

❝ Fed: replicate study

Right (that’s a ‘need to know’).* But EMA is interested in the comparison of T_fed vs. T_fasted and R_fed vs. R_fasted as well (that’s the ‘nice to know’ part).* If you run two 2×2 studies you can perform such a comparison only between these studies - which is lacking power.
Therefore EMA stated in the draft GL:

In cases where information is required in both the fed and fasted states, it is preferable to conduct a four-period single dose crossover design study (both products fed and fasted) rather than conducting two separate bioequivalence studies in fed and fasted state, respectively. In a four-period crossover design study, the food effect on test and reference product can be evaluated which is not the case when conducting two separate two-period, two-sequence single dose crossover design studies under fasting and fed conditions, respectively. In addition to the bioequivalence evaluation of test/reference in fasting and in fed state, the food effect can be presented for test and reference, i.e. the ratio food/fasting and 90% confidence interval for test and reference, respectively.

(my emphases) Of course the text in red in nonsense – such a comparison is just based on a parallel design. If the usual precautions (polymorphism, etc.) are followed in designing these studies, unbiased estimates and CIs may be obtained.

Following comments (p. 78-83) on the draft this paragraph was simplified; the final GL states only:

In cases where information is required in both the fed and fasted states, it is acceptable to conduct either two separate two-way cross-over studies or a four-way cross-over study.

(my emphases)

These terms where extensively discussed at the BioInternational Conference 1994 in Munich and are used ever since in that context.

—
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

Dr_Dan
★★

Germany,
2010-03-12 09:49
(5535 d 13:12 ago)

@ Helmut
Posting: # 4900
Views: 13,898

What if…

Post reply

Dear Helmut
You are a statistician, I am a project manager. The purpose of a bioequivalence study is not to describe the full pharmacokinetics (including food effect) of a drug entity but to detect differences between galenic formulations. Since it is acceptable to conduct either two separate two-way cross over-studies or a four-way cross-over study in cases where information is required in both the fed and fasted states I would always suggest to perform two separate two-way cross over-studies. I totally agree that from a scientific point of view the four-way cross-over study is much more interesting but from a practical (generic company's) point of view you should prefer the other option.
Kind regards
Dan

—
Kind regards and have a nice day
Dr_Dan

Helmut
★★★

Vienna, Austria,
2010-03-12 12:35
(5535 d 10:26 ago)

@ Dr_Dan
Posting: # 4905
Views: 13,889

Need-to-know / nice-to-know

Post reply

Dear Dan,

no, you got me wrong. ;-)

I was always opposing the 4-way study; reasons:

The variability in fed state may be different to fasted state. Since you have to power the study for the 'weakest' arm, the 4-way study may need more treatments than two 2×2 studies.
Confounded effects of formulation and food – biased estimates.
Higher blood loss, drop-out rate like in any other 4-way study.

❝ I totally agree that from a scientific point of view the four-way cross-over study is much more interesting but from a practical (generic company's) point of view you should prefer the other option.

Almost. I can understand why EMA suggested a 4-way study in the first place. If you perform the comparison between the fasted and fed studies, the confidence intervals will be quite large. The 4-way study would overcome this problem. But since it is only a 'nice to know' term, no more restrictions in the final GL.
But: I would expect that EMA will ask for such a comparison. So plan the two studies as if they are two treatment arms in a parallel design.

—
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

boonchai_l
☆

Thailand,
2010-05-24 13:53
(5462 d 10:08 ago)

@ Helmut
Posting: # 5368
Views: 13,641

many ANOVA but only one trial

Post reply

Dear All,

According to above recommendations for 4-way crossover study, I am a little bit weird because we were planning to conduct at least two ANOVA tables from only one trial.

This is a possible example.

The overall ANOVA may be produced to tell that there is any significance, the formulation may be a main effect and food is block effect, something like this... but any variance in this ANOVA is not suitable to conduct 90%CI to prove equivalence in both conditions. So
Some data in some periods and sequences are selected to produce ANOVA for obtaining the MSE to conduct 90%CI of T/R ratio under fed condition. and
Some data in some periods and sequences are selected to produce ANOVA for obtaining the MSE to conduct 90%CI of T/R ratio under fasting condition.

BTW: The major objective of most study for both of clinical or non-clinical experiment is to prove the difference among treatments. The F statistic can tell whether that difference is significant. And some contrast may be set to prove significance in some interesting condition.

Unlike bioequivalence study, it is for proving equivalence. So the F and ANOVA are not the answer to prove equivalence BUT the thing everyone needs to know is whether 90%CI is in 80-125%.

Best Regards,
BL

GSTATS
☆

India,
2010-06-08 23:52
(5447 d 00:09 ago)

(edited on 2010-06-09 16:01)
@ boonchai_l
Posting: # 5482
Views: 13,514

many ANOVA but only one trial

Post reply

Dear BL,

For four way crossover design with following sequences

    Tfasted  Rfasted  Tfed  Rfed

    Rfasted  Tfasted  Rfed  Tfed

We don't need to built many ANOVA. What you can do is add diet(fed or fast) as factor in your ANOVA and accordingly modify other factors.
Then using repeated statement in PROC mixed calculate residual variability for two diets from which intra subject variabilty under fed and fasted state can be calculated separately.

And for comparisons like Tfasted vs Rfasted, Tfed vs Rfed , use estimate statement which will provide estimated difference with confidence intervals. And for other comparisons like Tfasted vs Tfed etc. we can calculate ratios from LSMeans obtained through model.

But problem with this design, as HS said, is

❝ The variability in fed state may be different to fasted state. Since you have to power the study for the 'weakest' arm, the 4-way study may need more treatments than two 2×2 studies.

But design is good replacement for two or more studies if there is not much difference in variabilty of drug under fed and fasted state.

Regards,
GSTATS
http://www.gstatsolutions.com/Medical%20Research.html

—
Let Noble Thoughts come from Every Side: RIG VEDA

4-period BE study fed/fasting: statistical issues [General Sta­tis­tics]

Confounded effects; common variance

Neglecting period effects?

Assumptions...

What if...

What if...

What if…

What if…

Need-to-know / nice-to-know

many ANOVA but only one trial

many ANOVA but only one trial

4-period BE study fed/fasting: statistical issues [General Statistics]