Bioequivalence and Bioavailability Forum • Planing ABEL based on a pilot

Helmut
★★★

Vienna, Austria,
2013-09-18 18:27
(4655 d 14:58 ago)

Posting: # 11519
Views: 5,932

Planing ABEL based on a pilot [Power / Sample Size]

Dear all,

I stumbled across an interesting issue. Let’s say we plan for a replicate pilot study (RTR|TRT) in order to design the pivotal in the same design. I want to use the upper CL of the estimated CV (40%) by CVCL(CV=0.4, df=2*n-3, side="upper").

pilot n CL_upper 16 0.525 20 0.506 24 0.494 36 0.472

Sure. Now let’s plan the pivotal (T/R 90%, 80% power) for EMA’s method by sampleN.scABEL(theta0=0.9, CV=CV, design="2x2x3").

CV pivotal n power total n 0.525 44 0.8154 60 0.506 42 0.8021 62 0.494 42 0.8046 66 0.472 42 0.8046 78

In some cases we are already crossing the 50% cap. The larger pilot size does not pay off.
Let’s play the game for a CV of 30% (only 50% chance of scaling). If you prefer one-liners, use: sampleN.scABEL(theta0=0.9, CV=as.numeric(CVCL(CV=0.3, df=2*n-3, side="upper")[2]), design="2x2x3"). I got:

pilot n pivotal n power total n 16 46 0.8008 62 20 48 0.8058 68 24 48 0.8007 72 36 50 0.8046 86

Amazing. Small is beautiful?

—
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,
2013-09-23 11:24
(4650 d 22:01 ago)

@ Helmut
Posting: # 11542
Views: 4,710

Planing ABEL based on a pilot

Post reply

Dear Helmut,

❝ Amazing. Small is beautiful?

For me your finding is not so astonishing.
Remember one of the features of the scaled ABE method: In its pure form the power is independent of the intra-subject variability, at least if we assume a true ratio T/R of ~1. Therefore the FDA had discussed about a fixed sample size (24 or 36) for such studies.

The simplification of the scaled ABEL method and the additionally regulatory constraints like cap on widening the acceptance ranges or the PE constraints of course change that behaviour. But as your numbers show not to that extent that you observe a benefit for larger pilots.

I would nevertheless not state "Small is beautiful". This would only be the case if you take only the variability from the pilot. The other side of the moon is the PE and its value may be very misleading in small pilots, especially if we talk HVD.

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2013-10-26 19:37
(4617 d 13:49 ago)

@ d_labes
Posting: # 11777
Views: 4,796

Planing ABEL based on a pilot

Post reply

Dear Detlew & all,

❝ I would nevertheless not state "Small is beautiful".

Correct. I think I was getting the right answer to a wrong question. The CV’s CI helps us to be protected against surprises. In the conventional ABE setting that would mean working with the upper CL, but if scaling comes into play I would say the lower CL is important (lower CV, smaller widening of the AR, higher sample size). My revised example based on CVCL(CV=0.4, df=2*n-3, side="lower"):

pilot n CL_lower 16 0.326 20 0.333 24 0.338 36 0.348

And sampleN.scABEL(theta0=0.9, CV=CV, design="2x2x3").

CV pivotal n power total n 0.326 52 0.8082 68 0.333 52 0.8098 72 0.338 52 0.8113 78 0.348 50 0.8028 86

Again smaller pilot sample sizes result in smaller total ones, but your argument about a more precise estimate of the T/R-ratio in larger pilot studies is still valid.

—
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