Bioequivalence and Bioavailability Forum 02:01 UTC

Main page Policy/Terms of Use Abbreviations Latest Posts

 Log in |  Register |  Search

Replicate vs. 2×2×2 [Regulatives / Guidelines]

posted by Helmut Homepage - Vienna, Austria, 2017-05-15 17:07  - Posting: # 17355
Views: 14,486

Hi M.tareq,

» » while ago I reviewed a manuscript exploring the pros and cons of TSDs vs. ABEL. Was very interesting and I hope that the authors submit a revised MS soon.
»
» Can i get the link or the name of the paper when the authors submit it ?

If the revised MS will get accepted and published, for sure.

» In General, if a drug isn't known to be HVDP and by definition of HVDP that, they have wide therapeutic index yet the CRO/Sponsor went with replicate design.

Any replicate design can be assessed for conventional ABE. If the sponsor intends to try RSABE it has to be stated in the protocol. Might be worthwhile in borderline cases (CVwR ~30%).

» How the assessor would consider that? mean if the published literature or pilot study suggest low CV of the reference /test product yet the cro/sponsor went with replicate design?

I think you are mixing two things up. Replicate designs are applicable for ABE as well. If the sponsor aims for RSABE – in contrary to data in the public domain (few exist!) which suggests low variability – as a regulator I would be cautious. Note that regulators see a lot of studies. Was the CVwR caused by poor study conduct? That’s a gray zone.

Pilot studies can be misleading. The estimated CV is not carved in stone. Try this one:

library(PowerTOST)
design <- "2x2x4" # any of "2x2x4", "2x2x3, "2x3x3"
n   <- 16      # total sample size
CV  <- 0.25    # CV as a ratio
df  <- eval(parse(
         text=known.designs()[which(known.designs()$design==design), "df"],
         srcfile=NULL)
       )       # get the degrees of freedom
CL <- CVCL(CV=CV, df=df, side="2-sided")
cat("\ndesign of pilot:", design,
    "\nsample size    :", n,
    "\nestimated CV   :", sprintf("%.2f%%", 100*CV),
    "\nCL of the CV   :", sprintf("%.2f%% to %.2f%%", 100*CL[[1]], 100*CL[[2]]),
    "\n\n")

Gives:

design of pilot: 2x2x4
sample size    : 16
estimated CV   : 25.00%
CL of the CV   : 20.60% to 31.87%

Might be highly variable, right?

» Like abusing the use of scABE ? :confused:

Maybe.

» » Nobody knows how to deal with this story. :-( Regulations ≠ science. Not required by the FDA…
»
» Agree, yet i read an ANDA submission -can't find the link now- where the sponsor detected outliers based on T/R ratio of each subject and redosing of such subjects along with other subjects who exhibited normal pkp profile; though it was after reviewing with FDA regulator.

Yes, that’s an old story (“re-dose the suspected outliers – both with T and R – together with at least five ‘normal’ subjects or 20% of the sample size, whichever is larger”). IMHO, the individual T/R-ratios are not a particularly good idea for assessing outliers (ignoring period effects).

» Another published paper about ibandronic acid https://www.ncbi.nlm.nih.gov/pubmed/24756462

Highly variable like all bisposphonates. Can’t reproduce the estimated sample. I got 132 and not 138.

» … the sponsor/CRO stated the definition of outliers using studentized residuals and boxplot to eliminate subjects with values away from the boxplot by more than 3 IQR.

Not surprising since Susana Almeida was co-chair of the Bioequivalence Working Group, European Generic and Biosimilar Medicines Association (EGA). At the joint EGA/EMA workshop (London, June 2010) I – as a joke! – suggested boxplots, which are nonparametric by nature. The EMA hates nonparametric statistics. But the panelists welcomed the “idea” and my joke made it to the Q&A document. My fault. Never presume humor. Now it’s carved in stone.

This study demonstrated why not accepting reference-scaling for AUC might not be a good idea. Since the sample size is based on AUC, products with extreme T/R-ratios of Cmax would pass ABEL due to the high sample size. Given the results of this study: 90% chance to pass with T/R as low as 84.46% or 80% chance to pass with T/R 82.95%. Technically (i.e., according to the GL) nothing speaks against that. But do we want such products on the market?
Wasn’t the case in this study (T/R for Cmax 102.56%), but…
BTW, for the FDA the study could have been performed in just 45 (!) subjects – without risking extreme Cmax-ratios.

» My point is, as your kindly said, it's best to state in the protocol how to deal with outliers especially regarding variability estimation and proving/showing to the assessor the reasons for excluding such outlier(s) from study or reviewing it with the regulator before submission of the data?

I would say so.

Cheers,
Helmut Schütz
[image]

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

Complete thread:

Activity
 Mix view
Bioequivalence and Bioavailability Forum |  Admin contact
18,914 posts in 4,036 threads, 1,283 registered users;
online 17 (0 registered, 17 guests [including 14 identified bots]).

Statistics is, or should be, about scientific investigation
and how to do it better, but many statisticians believe
it is a branch of mathematics.    George E.P. Box

The BIOEQUIVALENCE / BIOAVAILABILITY FORUM is hosted by
BEBAC Ing. Helmut Schütz
HTML5 RSS Feed