Bioequivalence and Bioavailability Forum • Use of 96.7% CI instead of 90% CI

pjs
★

India,
2015-07-08 19:07
(3547 d 19:15 ago)

Posting: # 15053
Views: 4,166

Use of 96.7% CI instead of 90% CI [Regulatives / Guidelines]

Dear all,

In One of the UKPAR, applicant had used 96.7% of CI instead of 90% CI. The comment from the agency was:

The applicant has provided suitable justification for the use of the adjusted confidence interval 96.7 % instead of a confidence limit 90%.

What could be the reason to use 96.7% CI (decreasing the patient risk ??? increasing the faith in product???)

and what justification would have been given.

Regards,
PJS

ElMaestro
★★★

Denmark,
2015-07-08 19:29
(3547 d 18:53 ago)

@ pjs
Posting: # 15054
Views: 3,470

Use of 96.7% CI instead of 90% CI

Post reply

Hi pjs,

❝ What could be the reason to use 96.7% CI (decreasing the patient risk ??? increasing the faith in product???)

❝

❝ and what justification would have been given.

I have no actual idea, but I think some kind of multiplicity issue could have been the case here since they worked with a decreased alpha. Perhaps it was something with a semi-sequential design, multiple candidate formulations and/or interim picking of a final one for finalisation, or something along those lines?

—
Pass or fail!
ElMaestro

Helmut
★★★

Vienna, Austria,
2015-07-09 02:43
(3547 d 11:39 ago)

@ pjs
Posting: # 15056
Views: 3,662

Bonferroni (three tests)

Post reply

Hi PJS,

❝ The applicant has provided suitable justification for the use of the adjusted confidence interval 96.7 % instead of a confidence limit 90%.

❝

❝ What could be the reason to use 96.7% CI

Smells of Bonferroni’s adjustment for three simultaneous comparisons.

❝ (decreasing the patient risk ??? increasing the faith in product???)

Neither nor. To preserve the patient’s risk at ≤0.05.

unadjusted (nominal) adjusted for multipl. k α CI FWER α CI FWER 1 0.05 90.00 0.0500 0.0500 90.00 0.0500 2 0.05 90.00 0.0975 0.0250 95.00 0.0494 3 0.05 90.00 0.1426 0.0167 96.67 0.0492

k is the number of tests, α the level of the test, CI the confidence interval calculated as 100(1–2α), and FWER the familywise (type I) error rate according to 1–(1–α)^k.
If three tests are performed at 0.05 the FWER might increase to ~14%. With an adjusted level of 0.05/3=0.0166 (96.66% CI) the FWER is preserved at ~4.9%.

❝ and what justification would have been given.

The Bonferroni correction is in general considered the most conservative amongst multiplicity adjustment methods and therefore, requires the least (if any substantial) justification. Might be more demanding to convince regulators about using others (e.g., Šidák. Holm, Hochberg,…).
It is not evident from the PAR why the company used it (BE of a solution vs. an IR formulation).
I’ve applied it when showing BE at one dose level and in the same study assessed dose proportionality of the test. Example:

dose T R 10 ☒ ☐ 20 ☐ ☐ 40 ☒ ☐ 60 ☐ ☐ 80 ☒ ☒

BE of T vs. R at 80mg, dose proportionality of T shown at 10/40/80mg. Biowaivers for 20/60mg.

—
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

pjs
★

India,
2015-07-09 11:11
(3547 d 03:11 ago)

@ Helmut
Posting: # 15058
Views: 3,394

Bonferroni (three tests)

Post reply

Dear helmut,

Thanks for your reply.

❝ If three tests are performed at 0.05 the FWER might increase to ~14%.

I am not statistician, so i might be wrong. But as far as i understand if three test products are compared against the one reference product in a single study, then the study title shold be 4 treatment, four sequence (?). It has been two treatemnt, two sequence in the title.

Or they have removed the other test product results from the study and used the Bonferroni correction?

One article that helped understand some aspects is article

Thanks,
PJS

Helmut
★★★

Vienna, Austria,
2015-07-09 14:18
(3547 d 00:04 ago)

@ pjs
Posting: # 15059
Views: 3,489

Bonferroni (three tests)

Post reply

Hi PJS,

❝ I am not statistician, so i might be wrong. But as far as i understand if three test products are compared against the one reference product in a single study, then the study title shold be 4 treatment, four sequence (?).

I’m neither Sherlock Holmes nor do I own a crystal ball. I have no clue what was done in this study.

❝ It has been two treatemnt, two sequence in the title.

❝ Or they have removed the other test product results from the study and used the Bonferroni correction?

No idea. Concluding from the title to what was actually done might be misleading. I know one nitpicking Polish assessor regularly rejecting protocols for the title “A bioequivalence study…”. His argument is: The title should be “A comparative bioavailability study…”. Bioequivalence is the desired result.

I still hold that Bonferroni’s correction was applied because none of the other multiplicity corrections or adaptive methods (with interim analysis/es) would lead to such an α.

❝ One article that helped understand some aspects is article

THX for mentioning my friend Jiří Hofmann’s presentation! In the future please don’t use a 300+ character link pointing to -India but the URL of the document itself.

—
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