Bioequivalence and Bioavailability Forum • ANVISA’s home brewed scaling?

Helmut
★★★

Vienna, Austria,
2015-05-28 20:27
(3579 d 12:11 ago)

Posting: # 14877
Views: 11,613

ANVISA’s home brewed scaling? [RSABE / ABEL]

Dear all,

recently I saw a response to a request of one of my clients concerning reference-scaling:

“In consideration of your request, we clarify that the Anvisa has accepted the use of the statistical method for scaling proposed by the European Agency (EMA, 2010), with the exception that for Anvisa scaling can only be applied […] in the cases that the intra-subject coefficient of variation (CV%) exceeds 40% for the originator product.
Please be advised, however, that the clinical protocol must be previously submitted to the technical evaluation of Therapeutic Equivalence Coordination (Anvisa) before conducting the bioequivalence study that you want to consider the scaling.”

Interesting. As usual the devil is in the details. IMHO, two interpretations are possible.

Use EMA’s regulatory constant k 0.760 which is derived from the regulatory standard deviation σ₀ based on the switching condition CV_WR of 30%:
σ₀ = √ln(0.30²+1) = 0.2936
k = ln(1.25)/σ₀ ≈ 0.760
Scale according to EMA’s formula [L, U] = ℯ^±k·s_WR but apply widening of the acceptance range only if CV_WR >40%. This would lead to a discontinuity (similar to FDA’s “implied BE limits” based on a switching condition of ~25.4% [σ₀ 0.25] applied at ≥30%):
CV_WR AR (%) 0.20 80.00 125.00 0.25 80.00 125.00 0.30 80.00 125.00 0.35 80.00 125.00 0.40 80.00 125.00 0.45 72.15 138.59 0.50 69.84 143.19 0.55 69.84 143.19 0.60 69.84 143.19
If ANVISA wants 40% as the switching condition that would translate into σ₀ 0.3853 and k 0.579. No discontinuity but very little gain in terms of the acceptance range:
CV_WR AR (%) 0.20 80.00 125.00 0.25 80.00 125.00 0.30 80.00 125.00 0.35 80.00 125.00 0.40 80.00 125.00 0.45 77.99 128.23 0.50 76.07 131.46 0.55 76.07 131.46 0.60 76.07 131.46

Any experiences from our members? In any case, the sample size estimation will be challenging…

—
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

ElMaestro ★★★ Denmark, 2015-05-28 21:10 (3579 d 11:28 ago) @ Helmut Posting: # 14878 Views: 9,869	ANVISA’s home brewed scaling? Post reply
	Hi Helmut, ❝ […] in the cases that the intra-subject coefficient of variation (CV%) exceeds 40% for the originator product. Haha, for a moment I thought you were serious. But then realised that of course no agency will be enforcing such a weird and unjustified requirement. You almost had me there. Great sense of humour. — Pass or fail! ElMaestro

Helmut
★★★

Vienna, Austria,
2015-06-02 18:00
(3574 d 14:37 ago)

@ ElMaestro
Posting: # 14896
Views: 9,786

ANVISA: TIE, sample size

Post reply

Hi ElMaestro,

❝ […] But then realised that of course no agency will be enforcing such a weird and unjustified requirement.

No? IMHO, it is not more unjustified than EMA’s. The EMA had no negative results from arbitrarily widening the AR to 0.70~1.43 (pre-2006, 2×2 crossover sufficient; widening also for AUC!) or to 0.75~1.33 (2006+, replicate design, C_max only). There are hundreds (?) of products authorized according to those requirements. This “evidence” explains EMA’s cap on the CV_WR of 50%. They didn’t want to reach beyond what they have accepted before.
On the contrary FDA’s scaling came out of the blue… For CVs > ~50% the restriction on the PE of 0.80~1.25 is more important than the CI. Since in Canada (1991+) only the PE of C_max has to lie with 0.80~1.25 there is some evidence across the pond that it “works” as well. Experience with AUC? Nada.

Reading ANVISA’s response over and over again, I gather that they mean #1 in order to be on the safe side (“EMA accepts reference-scaling for five years now. Let’s be slightly more conservative.”). At least some of the inflation of the Type I Error observed with EMA’s “method” vanishes (I expect still a TIE of ~0.06 at 40%). Until Detlew releases an (experimental?) update for PowerTOST’s *.scABEL functions (regulator="ANVISA") one can only assume the worst in sample size estimation, i.e., unscaled ABE, theta0=0.9, CV=0.4…

—
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,
2015-06-03 13:59
(3573 d 18:39 ago)

@ Helmut
Posting: # 14899
Views: 9,577

ANVISA: scABEL TIE

Post reply

Dear Helmut!

❝ Reading ANVISA’s response over and over again, I gather that they mean #1 in order to be on the safe side (“EMA accepts reference-scaling for five years now. Let’s be slightly more conservative.”). At least some of the inflation of the Type I Error observed with EMA’s “method” vanishes (I expect still a TIE of ~0.06 at 40%). Until Detlew releases an (experimental?) update for PowerTOST’s *.scABEL functions (regulator="ANVISA") one can only assume the worst in sample size estimation, i.e., unscaled ABE, theta0=0.9, CV=0.4…

PowerTOST V1.2-07 is on the way.
First impression concerning TIE (n=24, nsims=1E6):

    CV   theta0      TIE

 0.250  1.250000 0.050508

 0.275  1.250000 0.050839

 0.300  1.250000 0.053183

 0.325  1.250000 0.061220

 0.350  1.250000 0.078676

 0.375  1.250000 0.105438

 0.400  1.250000 0.137350

 0.4001 1.340322 0.039371

 0.425  1.363072 0.039987

 0.450  1.385991 0.038561

 0.475  1.408971 0.035537

 0.500  1.431997 0.031668

 0.525  1.431997 0.036341

 0.550  1.431997 0.040046

 0.575  1.431997 0.042868

 0.600  1.431997 0.044908

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2015-06-03 14:07
(3573 d 18:30 ago)

@ d_labes
Posting: # 14900
Views: 9,599

ANVISA: scABEL TIE

Post reply

Dear Detlew!

❝ PowerTOST V1.2-07 is on the way.

THX!

❝ First impression concerning TIE (n=24, nsims=1E6):

❝ CV theta0 TIE

❝ 0.400 1.250000 0.137350

Terrible! In their attempt to be more conservative they are shooting themselves in the foot. The same story over and over again. :no:

—
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,
2015-06-03 15:19
(3573 d 17:18 ago)

@ Helmut
Posting: # 14901
Views: 9,588

ANVISA: scABEL TIE part II

Post reply

Dear Helmut!

❝ ❝ PowerTOST V1.2-07 is on the way.

❝ THX!

You are welcome. But wait until it arrives really.

❝ ❝ First impression concerning TIE (n=24, nsims=1E6):

❝ ❝ CV theta0 TIE

❝ ❝ 0.400 1.250000 0.137350

❝

❝ Terrible! In their attempt to be more conservative they are shooting themselves in the foot. The same story over and over again. :no:

Exactly.

But the picture changes if they claim the same hokus pokus like "consumer risk model" and "implementation" as in the paper
Davit et al.
"Implementation of a Reference-Scaled Average Bioequivalence Approach for Highly Variable Generic Drug Products by the US Food and Drug Administration"
AAPS Journal, Vol. 14, No. 4, December 2012

p(BE) at the original EMA widened acceptance limits, again n=24, design="2x3x3" (partial replicate), nsims=1E6:

     CV   theta0    p(BE)

 0.2500 1.250000 0.050508

 0.2750 1.250000 0.050839

 0.3000 1.250000 0.053183

 0.3250 1.272300 0.039390

 0.3500 1.294796 0.036086

 0.3750 1.317424 0.037143

 0.4000 1.340165 0.039410

 0.4001 1.340256 0.039408

 0.4250 1.363001 0.040021

 0.4500 1.385915 0.038601

 0.4750 1.408890 0.035578

 0.5000 1.431910 0.031708

 0.5250 1.431910 0.036390

 0.5500 1.431910 0.040094

 0.5750 1.431910 0.042911

 0.6000 1.431910 0.044961

—
Regards,

Detlew

nobody nothing 2015-06-03 15:31 (3573 d 17:06 ago) @ d_labes Posting: # 14902 Views: 9,505	ANVISA: scABEL TIE part II Post reply
	... Would you mind elaborating a little on the hokus pokus part of the story? — Kindest regards, nobody

d_labes ★★★ Berlin, Germany, 2015-06-03 16:00 (3573 d 16:37 ago) @ nobody Posting: # 14903 Views: 9,461	ANVISA: scABEL TIE part II Post reply
	Dear nobody, we had it already here to some degree of detailedness. Hope this is sufficient. — Regards, Detlew

d_labes
★★★

Berlin, Germany,
2015-06-05 13:03
(3571 d 19:34 ago)

@ Helmut
Posting: # 14916
Views: 9,430

ANVISA scABEL - sample size

Post reply

Dear Helmut! Dear All!

Proud to announce:
PowerTOST V1.2-07 is on CRAN now :cool:

.
After fiddling with such ugly requests like "Title case".

First look on sample size:
Settings: design="2x3x3", theta0=0.9 (acc. to the two Laszlo's), targetpower=0.8

 CV     N(EMA) N(ANVISA)

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

0.25      42       42

0.275     48       51

0.3       54       60

0.325     51       69

0.35      48       78

0.375     45       78

0.4       42       63

0.425     42       51

0.45      39       45

0.475     39       42

0.5       39       42

0.525     42       42

0.55      42       42

0.575     45       45

0.6       48       48

0.65      54       54

0.7       60       60

0.8       75       75

As you see there is a partly considerable rise in burden for the ANVISA settings compared to the EMA in the range CV=30 - 45%(!). In this range ANVISA is actually more conservative than EMA. Note the range CV=40-45% where the widened acceptance ranges are identical for both regulatory settings but sample size is higher for ANVISA.

After reaching the CV where a cap has to be placed on the widening (CV=50%) of the acceptance range both regulatory settings give the same sample size needed.

—
Regards,

Detlew

Lucas
★

Brazil,
2015-06-02 18:02
(3574 d 14:35 ago)

@ Helmut
Posting: # 14897
Views: 9,618

ANVISA’s home brewed scaling?

Post reply

Hello guys.

Very interesting indeed, huh? We've also received that kind of feedback, it's a standard response when ANVISA is questioned about BE approaches for HVDs. From what I heard here, the first approach described by HS up here is the one that they want. Same regulatory constant.
They think that drugs with variability between 30% and 40% are not HVDs in fact, since the rate of BE approval in this range is acceptable, according to them. IMHO that affirmation is reasonable, but how would that be used together with EMA's ABEL I wouldn't know how.

For sure planning a study in those conditions is very hard. Very few people have tried scaling for ANVISA, since it's not yet very well established how it should be done.

Hope to have been of help. Best regards.

Lucas

Helmut
★★★

Vienna, Austria,
2015-06-02 18:16
(3574 d 14:21 ago)

@ Lucas
Posting: # 14898
Views: 9,622

ANVISA’s home brewed scaling?

Post reply

Hi Lucas,

❝ […] the first approach described by HS up here is the one that they want. Same regulatory constant.

THX for clarifying!

❝ […] but how would that be used together with EMA's ABEL I wouldn't know how.

Technically no problem. See EMA’s SAS-code given for the “crippled models” (Methods A and B) in the Q&A-document. If the reference’s CV is not >40%, use the conventional acceptance range. Of course this leads to the weird discontinuity at 40%:

CV_WR s_WR L ~ U 0.400000 0.385253 80.00~125.00 0.400001 0.385254 74.62~134.02

❝ Hope to have been of help.

Absolutely – THX!

—
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