Bioequivalence and Bioavailability Forum

vasup ☆ India, 2013-07-29 10:37 (4710 d 07:02 ago) Posting: # 11093 Views: 10,185	Sample size [Regulatives / Guidelines] Post reply
	Hi Can any one clarify on "is there any regulatory guidance or barrier to use the more number of subjects than required". For example based on the calculation and to get required power 50 subjects are sufficient what will be fate of the study if i use 100 subjects. Is acceptable to regulatory agencies.

ElMaestro
★★★

Denmark,
2013-07-29 12:00
(4710 d 05:39 ago)

@ vasup
Posting: # 11096
Views: 9,348

Sample size

Post reply

Hi vasup,

❝ For example based on the calculation and to get required power 50 subjects are sufficient what will be fate of the study if i use 100 subjects. Is acceptable to regulatory agencies.

Well, although your power goes up you might be unnecessarily exposing some subjects to the IMPs and this could be seen as problematic in terms of ethics. Also, you may see that some of your CIs become so narrow that 1.00 isn't included; this in itself is not too much of a regulatory problem unless you read Danish guidelines, but in practice some people seem to have a mental blockade about it. If I get it right promotional material mentioning actual CIs is said to be more efficient if the CI spans over 1.00 than if it doesn't.

—
Pass or fail!
ElMaestro

jag009
★★★

NJ,
2013-07-29 17:57
(4709 d 23:42 ago)

@ vasup
Posting: # 11112
Views: 9,284

Sample size

Post reply

Hi,

❝ Can any one clarify on "is there any regulatory guidance or barrier to use the more number of subjects than required". For example based on the calculation and to get required power 50 subjects are sufficient what will be fate of the study if i use 100 subjects. Is acceptable to regulatory agencies.

You can have a larger sample size than the number suggested by the power calculation to compensate for potential dropouts due to length of the study and potential adverse events. However, you can't go too large a number.

100 subjects vs 50 subjects is a BAD example...

John

Helmut
★★★

Vienna, Austria,
2013-07-30 17:51
(4708 d 23:48 ago)

@ vasup
Posting: # 11125
Views: 9,614

Overpowering

Post reply

Hi vasup,

first I agree with what ElMeestro and John stated.

❝ […] use the more number of subjects than required".

My emphasis. The crucial point here is the term “required”. Some guidelines suggest 80–90% power (US-FDA, Canada-HPFB/TGD, Japan-NIHS, Brazil-ANVISA,…), whereas EMA uses the ambiguous phrase “appropriate”.

❝ For example based on the calculation and to get required power 50 subjects are sufficient what will be fate of the study if i use 100 subjects.

Let’s assume you want 90% power with expected CV of 29% and a ratio of 0.95. 50 subjects will give you 90.8% power. If your run the study in 100 – your assumptions are correct and you have no drop-outs – power will be 99.5%. On the other hand, it allows to still have >80% power if the ratio drops to 0.89. The main problem will be IEC-approval. The IEC is concerned with the safety of subjects in the study. Why should they approve a protocol with such a high sample size only minimizing the producer’s risk?

❝ Is acceptable to regulatory agencies.

Some issues:

Submitting of the protocol: At least agencies with a statement about power (≤90%), might reject the protocol. They might also be concerned about “forced bioequivalence”, i.e., although you expect to have a “bad” ratio but don’t officially (!) base your sample size on it (some kind of camouflage tactic).
If the protocol was approved by the agency + IEC and the study was performed in 100 subjects other issues enter the scene:
- The PE is 0.95 (or even closer to 1 and/or lower CV) the study is extremely overpowered. Given that post hoc (aka retrospective) power is not part of the bioequivalence decision, there should be no problem with acceptance. You only wasted money and dosed more subjects than necessary – an ethically problematic situation you should avoid in the future.
- The PE is 0.89* (or even lower if the CV is lower as well) technically you have demonstrated BE. With a PE of 0.85 and a CV of 26% you still have a chance of 51% to show BE. Now put yourself in an assessor’s position: Imagine they have already approved other products with differences of <±5% to the reference in <50 subjects. Assessors decide on the whole body of evidence and “guidelines are guidelines are guidelines”. Do you think they will approve a product with a ratio 10% worse than the other generics on the market? I would not take that for granted. You might end up in a referral and a lot of nasty legal stuff with rather unpredictable outcome.

Confirming what you secretly expected but didn’t state in the protocol because it looks so awful.

—
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,
2013-07-30 18:37
(4708 d 23:02 ago)

@ Helmut
Posting: # 11127
Views: 9,413

Overpowering

Post reply

Hi all,

❝ My emphasis. The crucial point here is the term “required”. Some guidelines suggest 80–90% power (US-FDA, Canada-HPFB/TGD, Japan-NIHS, Brazil-ANVISA,…), whereas EMA uses the ambiguous phrase “appropriate”.

And furthermore things are complicated by the fact that we need Cmax and AUCt (and sometimes other metrics) to show BE, but when people dimension their studies they usually power things as if just one metric needs to be BE.

Example: We expect T/R 95% (yes, geometric mean on the observed scale :-D

) and CV=24% (observed scale :-D

) for AUCt and we expect T/R95% and CV 28% for Cmax.
Since Cmax is the worst case, we think we achieve 80% power with 34 evaluable subjects assuming the usual statistical stuff and with the implicit assumption that if Cmax is BE then AUCt will be as well. That works reasonably well in many cases, especially SODFs, because AUCt and Cmax tend to be positively correlated in the sense that if Cmax goes up then, all other factors being equal (which in itself can be a stretch), AUCt goes up as well.
Unfortunately the strength and mode of that correlation is in as good as all cases unknown. And for some formulations like DPIs it is outright nasty on the odd occasion.

In the absence of a good (as in valid and correct) model we therefore don't know the true power of any study even if the ordinary statistical assumptions hold. I'd like this topic to be better covered by stats literature...
To relate this boring story to the poster's questions I think we often take it for granted that if the worst-case is BE then the other will be as well. And in cases where we for some reason or other feel insecure about it, we could go for a higher (apparent; for one metric) power but we'd have no way of quantifying what 'higher' really should be or how to make 'higher' correspond to 'appropriate'.

This is real life. There should be more statisticians among CSO's in generic companies.

—
Pass or fail!
ElMaestro

jag009 ★★★ NJ, 2013-07-30 18:45 (4708 d 22:54 ago) @ ElMaestro Posting: # 11128 Views: 9,324	Overpowering Post reply
	Hi ElMaestro, ❝ This is real life. There should be more statisticians among CSO's in generic companies. And "they" as in company management or others sometime would blame us "Oh you didn't use enough subjects" because some believe that we are hired to make the numbers look good My answers were often like "What can I do if your formulation looks like poop" Jon

ElMaestro
★★★

Denmark,
2013-07-30 18:51
(4708 d 22:47 ago)

@ jag009
Posting: # 11129
Views: 9,333

Overpowering

Post reply

Ho John,

❝ My answers were often like "What can I do if your formulation looks like poop"

With diplomatic skills like that I imagine you might have had an ultra short half-life in the pharma industry :-D

Of course you are right. With OIPs we sometimes see things like
T/R for Cmax=140%
T/R for AUCt=85%

-and company big wigs expect their formulation staff to come up with quickfixes.

These things are formulations where there are really no well-described pharmaceutical technologies available to tweak the two independently such as is the case with many SODFs. Poop indeed.

—
Pass or fail!
ElMaestro

jag009 ★★★ NJ, 2013-08-01 22:08 (4706 d 19:31 ago) @ ElMaestro Posting: # 11153 Views: 9,171	Overpowering Post reply
	❝ ❝ My answers were often like "What can I do if your formulation looks like poop" ❝ ❝ With diplomatic skills like that I imagine you might have had an ultra short half-life in the pharma industry Been here since graduation!!! John

kumarnaidu
★

Mumbai, India,
2013-09-02 14:22
(4675 d 03:17 ago)

@ jag009
Posting: # 11405
Views: 9,036

Overpowering

Post reply

Hi,
We have conducted two fed studies for a particular molecule, first one was failed.

Study 1 details N=42
Cmax ratio: 117.71 (109.62, 126.40) with CV: 19.56

Study 2 details N=52
Cmax ratio: 106.29 (98.12, 115.15) with CV: 24.64

From earlier published studies we found that they have conducted the study for N=30 subjects with CV: 33 and the study was passed but they are not talking about the power of the study.

Now we want to conduct another study for the same molecule then should we go with the study 2 conducted by us or with the literature with high CV of 33%.

Helmut
★★★

Vienna, Austria,
2013-09-02 14:42
(4675 d 02:57 ago)

@ kumarnaidu
Posting: # 11406
Views: 9,044

Conservative study planning

Post reply

Hi Kumar,

❝ We have conducted two fed studies for a particular molecule,…

More important: same formulation?

❝ From earlier published studies we found that they have conducted the study for N=30 subjects with CV: 33 and the study was passed but they are not talking about the power of the study.

“Power of the study” is meaningless. It might be that they planned for a larger deviation of test from reference and lower CV than what was later observed in the study. In order words, the study passed more or less by chance (i.e., with a higher producer’s risk).

❝ Now we want to conduct another study for the same molecule then should we go with the study 2 conducted by us or with the literature with high CV of 33%.

I would suggest to base study planning on your own results (you have little information from the literature about side conditions which have an impact on the variability: food, posture, analytical method, …). I would also suggest to be conservative and use the upper confidence limit of the CV, which for the second study is 29.7%.

—
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

kumarnaidu
★

Mumbai, India,
2013-09-02 16:36
(4675 d 01:03 ago)

@ Helmut
Posting: # 11408
Views: 8,950

Conservative study planning

Post reply

Thanks Helmut,
The formulation is same.
One more thing I want to ask, by using mean ratio: 1.05, CV: 0.297 and power: 0.8 I am getting sample size of around 40. But as we know we failed to prove bioequivalent in our first study with sample size of 42 subjects then should we remain stick to 40 or some more research needed in this. Also earlier studies for the same formulation by others is showing that 30 to 40 is enough. Please guide me.

Ohlbe
★★★

France,
2013-09-02 20:44
(4674 d 20:55 ago)

@ kumarnaidu
Posting: # 11410
Views: 8,960

Conservative study planning

Post reply

Dear Kumar,

❝ Also earlier studies for the same formulation by others is showing that 30 to 40 is enough.

Same molecule, not same formulation ! Unless you copied theirs ? Or do you mean other studies of your formulation at another CRO ?

And what point estimate did they get ? Yours was 117.71 in your first study, no wonder the study failed. In the second the ratio was 106.29, which is much better but still higher than the 1.05 that you used in your calculations. What are you keeping your fingers crossed for ?

—
Regards
Ohlbe

Jay ☆ India, 2013-09-06 15:55 (4671 d 01:44 ago) @ Helmut Posting: # 11450 Views: 8,818	confidence limit of the CV Post reply
	Dear Helmut, ❝ I would also suggest to be conservative and use the upper confidence limit of the CV, which for the second study is 29.7%. Will you kindly guide in calculation of upper confidence limit of the CV.

Helmut
★★★

Vienna, Austria,
2013-09-06 17:02
(4671 d 00:37 ago)

@ Jay
Posting: # 11451
Views: 8,858

CL of CV

Post reply

Hi Jay,

❝ […] guide in calculation of upper confidence limit of the CV.

I would suggest to get package PowerTOST for R. Easy with Kumar’s data (CV_w 24.64% in 52 subjects):

require(PowerTOST) CVCL(CV=0.2464, df=52-2, side="upper")

which gives:

lower CL upper CL 0.0000000 0.2974387

If you want to walk the stony path of manual calculation:

\({s_{w}}^{2}=\log{({CV_{w}}^{2}+1)}=\log{(0.2464^2+1)}=0.05894\).
In a 2×2 cross-over where n is the total number of subjects: ν = n – 2
\(\alpha=0.05 \to \chi_{\alpha,\nu}^{2}=34.764\).
\(CL_{upper}=\sqrt{e^{{\sigma_{w}}^{2}\cdot\nu/\chi_{\alpha,\nu}^{2}}-1}=\sqrt{e^{0.05894\times50/34.764}-1}={\color{DarkGreen}{0.2974\sim29.7\%}}\).

—
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

Jay ☆ India, 2013-09-06 17:21 (4671 d 00:18 ago) @ Helmut Posting: # 11452 Views: 8,780	CL of CV Post reply
	Thanks Helmut. Is this method i.e. Upper CI of CV, should be considered while calculating sample size everytime? We use highest CV of Cmax, Auc0-t or AUCinf observed in pilot or from any literature. And is this use of upper CI of CV in sample size calculation commonly used in BE Studies? As much unawareness about its usage in Sample size calculation is not here.

Helmut
★★★

Vienna, Austria,
2013-09-06 17:42
(4670 d 23:57 ago)

@ Jay
Posting: # 11453
Views: 8,980

CL of CV

Post reply

Hi Jay,

❝ Is this method i.e. Upper CI of CV, should be considered while calculating sample size everytime?

I would say so – unless you have a bunch of studies with similar CVs.

❝ We use highest CV of Cmax, Auc0-t or AUCinf observed in pilot or from any literature.

Remember the CV is an estimate itself which carries some degree of uncertainty. This value is not set in stone.
BTW, you can easily see from the formula that the CV from a smaller study is more uncertain (i.e., higher CL due to the lower degrees of freedom) than the one from a larger study. Therefore, the price you have to pay (= sample size penalty if you use the CV’s upper CL) is higher if the pilot study was small.

❝ And is this use of upper CI of CV in sample size calculation commonly used in BE Studies?

Regrettably not. A sensitivity analysis in sample size planning is suggested in ICH-E9 (Section 3.5). Sponsors and CROs don’t care* and the former complain afterwards. :no:

❝ As much unawareness about its usage in Sample size calculation is not here.

–1 × –1 = +1?

Too bad. Imagine you (falsely) believe that the CV is some kind of a natural constant. Since it is an estimate, the true value lies somewhere within its confidence interval. If you plan a study with the “carved from stone” CV you have a 50% chance that the value in the study will be higher. Maybe you fail to show BE (sample size too low). If the value is lower – though you loose money (sample size higher than necessary) – you still show BE. Bingo.

— Sponsors have a tendency to be overly optimistic (Our product is great! Why should we expect a higher CV? Our ratio will be 1 – for sure!) and are always looking for the cheapest “solution”, forgetting that repeating a failed study is not very economic.
— CROs in secret love failed studies (Hey, we might get another – even larger – one!)…

Both views are ethically questionable and bad statistical practice as well (ICH-E9: The number of subjects in a clinical trial should always be large enough to provide a reliable answer to the questions addressed.)

—
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

Jay ☆ India, 2013-09-11 13:02 (4666 d 04:37 ago) @ Helmut Posting: # 11474 Views: 8,742	PE consideration Post reply
	Dear Helmut, Can we consider CV from a range, i.e. we got 24.7 CV from pilot and literature reports as high as 30. So can we use Cv of 26 or 27 for sample size calculation? And also we got point estimate of 1.06. So can we use 1.06 PE in our sample size calculation. Can we also use PE from literature, we observed PE of 1.08. Or we have to stick to 0.95 to 1.05%?

Dr_Dan
★★

Germany,
2013-09-11 14:07
(4666 d 03:32 ago)

@ Jay
Posting: # 11475
Views: 8,602

PE consideration

Post reply

Dear Jay
what about the third parameter, the intended power?

Formulation difference  5%, CV% 25,   intended Power 80% ⇒ n=28

Formulation difference  5%, CV% 27.5, intended Power 80% ⇒ n=34

Formulation difference  5%, CV% 25,   intended Power 90% ⇒ n=38

Formulation difference 10%, CV% 25,   intended Power 80% ⇒ n=56

Formulation difference 10%, CV% 27.5, intended Power 90% ⇒ n=92

At the end you have to perform a risk/cost evaluation.
Kind regards
Dan

—
Kind regards and have a nice day
Dr_Dan

Jay ☆ India, 2013-09-11 15:09 (4666 d 02:30 ago) @ Dr_Dan Posting: # 11477 Views: 8,620	PE consideration Post reply
	Dear Dr_Dan, we have to stick to 5% and 10% Formulation difference? Can one use 1.06 or 1.08 PE? Intended power will be more than 80%.

kumarnaidu
★

Mumbai, India,
2013-09-11 15:22
(4666 d 02:17 ago)

(edited on 2013-09-12 06:45)
@ Dr_Dan
Posting: # 11478
Views: 8,768

PE consideration

Post reply

Dear Dan,
Can we use 10% formulation difference. I saw one CRO had calculated the sample size based on the mean ratio (T/R)=88% (from pilot study). Is it appropriate to consider this much difference. In FDA guidance "Statistical Approaches to Establishing Bioequivalence" they have written the below line in sample size consideration section.

Sample sizes for average BE should be obtained using published formulas. Sample sizes for population and individual BE should be based on simulated data. The simulations should be conducted using a default situation allowing the two formulations to vary as much as 5% in average BA with equal variances and certain magnitude of subject-by-formulation interaction. The study should have 80 or 90% power to conclude BE between these two formulations. Sample size also depends on the magnitude of variability and the design of the study. Variance estimates to determine the number of
subjects for a specific drug can be obtained from the biomedical literature and/or pilot studies.
Please guide me.

—
Kumar Naidu

Helmut
★★★

Vienna, Austria,
2013-09-11 16:13
(4666 d 01:26 ago)

@ kumarnaidu
Posting: # 11479
Views: 8,584

Sockpuppet?

Post reply

Dear Kumar (aka Jay?),

since both of you registered from the same IP-range and posted from the same IP I get the impression you are the same person – or one a sockpuppet of the other. This is bad style and against the Forum’s Policy.

If you don’t show within one week evidence that this assumption is wrong (please [image]

the admin) I will block your accounts.

2013-09-11 14:38 CEST: Evidence provided, case closed. [Helmut]

—
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

kumarnaidu ★ Mumbai, India, 2013-09-12 15:04 (4665 d 02:35 ago) @ kumarnaidu Posting: # 11483 Views: 8,551	PE consideration Post reply
	Hi all, Can anybody reply to the above query of PE. Thanks in advance Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post! Please comply with the Forum’s Policy. [Helmut] — Kumar Naidu

Sample size [Regulatives / Guidelines]

Sample size

Sample size

Overpowering

Overpowering

Overpowering

Overpowering

Overpowering

Overpowering

Conservative study planning

Conservative study planning

Conservative study planning

confidence limit of the CV

CL of CV

CL of CV

CL of CV

PE consideration

PE consideration

PE consideration

PE consideration

Sockpuppet?

PE consideration