Jump to content


Photo

“leave-one-out” geometric mean

outliers

  • Please log in to reply
3 replies to this topic

#1 Helmut Schütz

Helmut Schütz

    Advanced Member

  • Members
  • PipPipPip
  • 316 posts
  • LocationVienna, Austria

Posted 17 January 2018 - 11:24 AM

Dear all,

according to the EMA’s BE-GL Section 4.1.8:
 

Exclusion of data cannot be accepted on the basis of statistical analysis or for pharmacokinetic reasons alone, because it is impossible to distinguish the formulation effects from other effects influencing the pharmacokinetics.


The exceptions to this are:

  1. A subject with lack of any measurable concentrations or only very low plasma concentrations for reference medicinal product. A subject is considered to have very low plasma concentrations if its AUC is less than 5% of reference medicinal product geo­metric mean AUC (which should be calculated without inclusion of data from the outlying subject). The exclusion of data due to this reason will only be accepted in exceptional cases and may question the validity of the trial.
  2. Subjects with non-zero baseline concentrations > 5% of Cmax. Such data should be excluded from bioequivalence calculation […].

Setting up a workflow for #2 is straightforward. However, I have no clue for #1 (currently I’m doing it in R). Any suggestions appreciated.

Bonus question: What about replicate designs where data might be incomplete (i.e., missing periods)?


 Best regards,
Helmut

https://forum.bebac.at/


#2 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 17 January 2018 - 11:37 AM

Hi Helmut, this ties in rather nicely with Colby's question yesterday;

https://support.cert...ices/#entry4987

 

I assume you want ot have a workflow that does this 'automatically'?  I think it can be done by using flags set either by the new features in NCA of version 8 or a datawizard to find the low exposure subject.

 

Then merging a flag back onto the raw data before filtering and creating the 'final' NCA onthat.  I think it might be a bit complex but I am sure it should be do-able.  Care to share an exemplar project?

 

 Simon.



#3 Helmut Schütz

Helmut Schütz

    Advanced Member

  • Members
  • PipPipPip
  • 316 posts
  • LocationVienna, Austria

Posted 17 January 2018 - 11:49 AM

Hi Simon,

 

Hi Helmut, this ties in rather nicely with Colby's question yesterday; https://support.cert...ices/#entry4987


I saw it. Interesting list. Will try to answer – time allowing.

 

I assume you want ot have a workflow that does this 'automatically'?


Yep. ;)

 

I think it can be done by using flags set either by the new features in NCA of version 8 or a datawizard to find the low exposure subject.
Then merging a flag back onto the raw data before filtering and creating the 'final' NCA onthat. I think it might be a bit complex but I am sure it should be do-able.


I was thinking along these lines. What worries me that for n subjects I need n–1 descriptive statistics to get the respective geometric means.

 

Care to share an exemplar project?


Not yet. You can play around with one of the example data sets (f.i. “Data 2x2.CSV”, “Data 2x4.CSV”).


 Best regards,
Helmut

https://forum.bebac.at/


#4 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 19 November 2019 - 02:09 PM

Ded Moroz has come early....

    This
solution coded in the data wizard will work with NAs, what do you think Helmut.

 

(note it is feasible there could be more than one PK metric for the subjects (i.e. replicate designs), that’s why the geo. mean is calculated for separate subjects too even if it does not seem to be necessary in this case.

 

  Simon.

 

Attached Files


  • bwendt@certara.com likes this




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users