Bioequivalence and Bioavailability Forum • 3x3 or 2x4?

felipeberlinski
☆

Brazil,
2014-10-22 16:20
(3856 d 19:41 ago)

Posting: # 13763
Views: 16,216

3x3 or 2x4? [RSABE / ABEL]

Dear friends

Reading a BE reccomendation of FDA I would like to know about your experience using two different designs.

Since the drug is a HVD:
Their reccomendation is to use a 3-sequence 3-period or a 2-sequence 4 period design.

Could anybody explain the main differences between them and when should I use each one? Go easy, I'm not a biostatistician :cool:

Thanks in advance

Edit: Category changed. [Helmut]

ElMaestro
★★★

Denmark,
2014-10-22 16:42
(3856 d 19:20 ago)

@ felipeberlinski
Posting: # 13764
Views: 14,561

3x3 or 2x4?

Post reply

Hi Felipe,

❝ Since the drug is a HVD:

❝ Their reccomendation is to use a 3-sequence 3-period or a 2-sequence 4 period design.

❝

❝ Could anybody explain the main differences between them and when should I use each one? Go easy, I'm not a biostatistician :cool:

When you do a replicated study for the purpose of proving BE by means of reference scaling it is of primary interest to you to have two administrations on Ref. and one administration of Test within each subjects.
Thus it isn't necessary to have two administrations of Test within each subject. Therefore I would personally prefer a 3-period, 3-sequence study. That designs is also lighter on the subjects - they undergo treatment (exposure) only in three periods and not four and that is arguably an ethical advantage.

I'll say this a little differently (principle applies to both the EU and US version, although there are subtle other difference in how it works under the hood, especially since the CI is implicit in the US). The three period design gives you:

The treatment effects of Test and Ref. You use this to derive the point estimate.
The variability of the point estimate. This determines the width of your CI.
The within-subject variability associated with the reference product. This one determines the acceptance range for the confidence interval.
The four-period design gives you all of the above plus
The within-subject variability for the Test product. No-one asks for this and to the best of my knowledge there is no way to use that estimate towards proving BE.

—
Pass or fail!
ElMaestro

nobody nothing 2014-10-22 17:06 (3856 d 18:55 ago) @ ElMaestro Posting: # 13765 Views: 14,402	3x3 or 2x4? Post reply
	Don't forget: find someone to calculate that kind of stuff. FDA differntiates between "normal" HV drugs (progesterone guidance document) and narrow therapeutic index Hv dugs (warfarin document) iirc... — Kindest regards, nobody

felipeberlinski ☆ Brazil, 2014-10-22 17:36 (3856 d 18:25 ago) @ nobody Posting: # 13766 Views: 14,396	3x3 or 2x4? Post reply
	Thank you both for the help and explanation. I fully understood your point on exposing subjects, and agree with you regarding the ethical question. But talking a little about sample size, does a 3x3 is advantageous on this matter? Which one will require a smaller number of subjects involved? Thank you again

nobody nothing 2014-10-22 17:41 (3856 d 18:20 ago) @ felipeberlinski Posting: # 13767 Views: 14,389	3x3 or 2x4? Post reply
	...as a starter: http://bebac.at/lectures/Mumbai2013WS1.pdf page 17 ...but drop outs may increase if you have to dose one individual FOUR times, with all the wash-out and potential adverse events... — Kindest regards, nobody

d_labes
★★★

Berlin, Germany,
2014-10-22 18:12
(3856 d 17:49 ago)

@ felipeberlinski
Posting: # 13768
Views: 14,834

2x3x3 or 2x2x3 or 2x2x4?

Post reply

Dear Felipe,

❝ But talking a little about sample size, ...

Since your original question refers to FDA lets see:
(using R with package PowerTOST, true GMR=0.95, target power=80%, sample size for reference scaled ABE using scABE criterion according to progesterone guidance)

> library(PowerTOST)

> sampleN.RSABE(CV=0.4, design="2x3x3")



++++++++ Reference scaled ABE crit. +++++++++

           Sample size estimation

---------------------------------------------

Study design:  2x3x3 (partial replicate) 

log-transformed data (multiplicative model)

1e+05 studies for each step simulated.



alpha  = 0.05, target power = 0.8

CVw(T) = 0.4; CVw(R) = 0.4

Null (true) ratio = 0.95

ABE limits / PE constraints = 0.8 ... 1.25 

Regulatory settings: FDA 

- CVswitch =  0.3 

- Regulatory constant = 0.8925742 



Sample size search

 n     power

21   0.73975 

24   0.80864 



> sampleN.RSABE(CV=0.4, design="2x2x3")



++++++++ Reference scaled ABE crit. +++++++++

           Sample size estimation

---------------------------------------------

Study design:  2x2x3 (TRT|RTR) 

log-transformed data (multiplicative model)

1e+05 studies for each step simulated.



alpha  = 0.05, target power = 0.8

CVw(T) = 0.4; CVw(R) = 0.4

Null (true) ratio = 0.95

ABE limits / PE constraints = 0.8 ... 1.25 

Regulatory settings: FDA 

- CVswitch =  0.3 

- Regulatory constant = 0.8925742 



Sample size search

 n     power

20   0.66022 

22   0.71427 

24   0.76072 

26   0.79722 

28   0.83092 



> sampleN.RSABE(CV=0.4, design="2x2x4")



++++++++ Reference scaled ABE crit. +++++++++

           Sample size estimation

---------------------------------------------

Study design:  2x2x4 (full replicate) 

log-transformed data (multiplicative model)

1e+05 studies for each step simulated.



alpha  = 0.05, target power = 0.8

CVw(T) = 0.4; CVw(R) = 0.4

Null (true) ratio = 0.95

ABE limits / PE constraints = 0.8 ... 1.25 

Regulatory settings: FDA 

- CVswitch =  0.3 

- Regulatory constant = 0.8925742 



Sample size search

 n     power

14   0.68849 

16   0.76590 

18   0.82274

The number of measurements (concentration samples) are
2x3x3 = 72 * number of timepoints
2x2x3 = 84 * number of timepoints
2x2x4 = 72 * number of timepoints

The design 2x2x3 needs somewhat higher sample size than the partial replicate but has the advantage of yielding also an intra-subject variability for Test. But this is a "nice to have".

Hope this helps.

—
Regards,

Detlew

MGR ★ India, 2015-04-08 15:35 (3688 d 20:26 ago) @ d_labes Posting: # 14675 Views: 13,092	2x3x3 or 2x2x3 or 2x2x4? Post reply
	Dear d_Labes, ❝ (using R with package PowerTOST, true GMR=0.95, target power=80%, sample size for reference scaled ABE using scABE criterion according to progesterone guidance) As per the above, can we do the sample size in SAS software, if so can we use pairedmeans option in procpower or anything? Could you please help me in this regards? Thanks in advance. — Regards, MGR

d_labes
★★★

Berlin, Germany,
2015-04-08 18:06
(3688 d 17:56 ago)

@ MGR
Posting: # 14676
Views: 13,179

Sample size for scABE in SAS?

Post reply

Dear MGR,

❝ As per the above, can we do the sample size in SAS software, if so can we use pairedmeans option in procpower or anything?

Sorry

. You have to do the power calculations via simulations because the methods underlying the scaled ABE have no (simple or complicated) analytical solution. And there is no Proc in SAS available to do this out of the box.

In principle the simulations could be programmed in SAS, but IMHO this is a horrible task. Moreover the run-time expected is very slow.
For that reasons (ease of programming, run-time) I had chosen the language R to implement the power/sample size estimation for scaled ABE, also my daily job is struggling with the beast SAS.

R is free available at CRAN, package PowerTOST also.
Both are relative easy to install. See this post.

Thus I recommend you to use that way to be able to do the sample size estimation for scaled ABE.

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2015-04-08 18:19
(3688 d 17:42 ago)

@ MGR
Posting: # 14677
Views: 13,073

Re-inventing the wheel

Post reply

Hi MGR,

adding to what Detlew said: See the paper* by the two Lászlós about simulations (done in MatLab). When asked for their code they recommended PowerTOST instead. ;-)

Endrényi L, Tóthfalusi L. Sample Sizes for Designing Bioequivalence Studies for Highly Variable Drugs. J Pharm Pharmaceut Sci. 2011;15(1):73–84. free resource.

—
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,
2014-10-22 18:36
(3856 d 17:25 ago)

@ felipeberlinski
Posting: # 13769
Views: 14,519

RTR|TRT

Post reply

Hi Felipe,

I oppose our Master when he wrote:

Thus it isn't necessary to have two administrations of Test within each subject.

Correct, if you have only CV_WR needed for scaling in mind. But – if CV_WR<30% – FDA’s mixed-effects code fails sometimes in partial replicated studies (TRR|RRT|RTR) since the model is over-specifed. Either parameterize the covariance structure as FA0(1) – instead of TYPE=FA0(2) given in the guidance – or perform a fully replicated three-period design (RTR|TRT). This design seemingly converges always and additionally gives you information about your product.

Sample sizes for FDA’s RSABE (T/R 90%, 80% power):

CV%          20  30  40  50  80

───────────────────────────────

RTRT|TRTR    32  24  24  22  28

TRR|RRT|RTR  30  45  33  30  42

RTR|TRT      28  46  38  34  44

I would not recommend a four-period study (more blood draws / subjects, higher drop-out rate).

—
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

Mauricio Sampaio
★

Brazil,
2014-10-28 13:12
(3850 d 21:50 ago)

@ felipeberlinski
Posting: # 13810
Views: 14,338

3x3 or 2x4?

Post reply

Hi Felipe! How are you my friend?

In my opinion, as the my focus is generics drugs products, I suggest the implementation of a reference scaled average bioequivalence using a 4x2 design. Because with test and reference formulations replicated the statistical adjustment is more consistent. Because you have more information from two medicines. In addition, the 4x2 design gives you a real information about your product that can be used in an internal discussion to improve your formulation in case of bioequivalence study fail.

More information read:

Implementation of a Reference Scaled Average Bioequivalence Approach for Highly Variable Generic Drug Products by the US Food and Drug Administration.
The AAPS Journal, Vol.14, No.4, December2012
doi 10.1208/s12248-012-9406-x

Best regards