Estimation of CVw and/or CVwR [Power / Sample Size]

posted by Helmut Homepage – Vienna, Austria, 2020-05-04 14:07 (1423 d 21:33 ago) – Posting: # 21393
Views: 7,843

Hi Alyssa,

Usually ISCV result of the 3 way crossover or 4 way crossover, replicate study published in the PAR is CVwR or CVw?


There are no rules (it depends on what the assessor decides to include of the study report). However, since the study was a replicate design with reference-scaling, possibly it is CVwR (more important).
We can estimate the CVw with the R-package PowerTOST:

library(PowerTOST)
round(100*CVfromCI(lower = 0.9625, upper = 1.2511,
                   n = 41, design = "3x6x3"), 1)

and get

Unbalanced 3x6x3 design. n(i)= 7/7/7/7/7/6 assumed.
[1] 36.8

Since this does not match what is given in the PAR, it is a strong hint that the 42.6% is the CVwR.
[Nonsense, not a Williams’ design! See ElMaestro’s post and the correction.]
If the expanded limits are given in the PAR, you can estimate the CVwR from the upper limit by the function CVwRfromU() to check. Example for 136.4%:

round(100 * CVwRfromU(136.4 / 100), 1)
[1] 42.6

Although CVwT is not accessible in a partial replicate design, CVw < CVwR means that the test is less variable than the reference (since CVw is pooled from CVwR and CVwT).

I agree with Dan but want to add one point. In my experience the variability across studies (with the same clinical setup, bioanalytical method, :blahblah:) tends to be more “stable” than the T/R-ratio. Hence, don’t fall into the trap of believing the nice 98.7% you observed in the pilot study. It might well have been pure chance. For HVD(P)s assuming a T/R-ratio of better than 90–111% is not a good idea (recommended by the two Lászlós* and therefore, the default in functions sampleN.scABEL() and sampleN.RSABE() of PowerTOST).


PS: Avoid sample size “calculation” if you don’t mind. Use “estimation” instead. ;-)
PPS: The partial replicate is a lousy design. If you want to have only three periods I suggest the 2×2×3 full replicate TRT|RTR instead. If you insist in the partial replicate (why?), use the function sampleN.scABEL.sdsims(). Slower than sampleN.scABEL() but more accurate. For a comparison see the vignette and scroll down to “Heterogenicity”.


Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
90 visitors (0 registered, 90 guests [including 9 identified bots]).
Forum time: 10:40 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5