Jump to content


Photo

samples under LOQ - handling in Phoenix cf NONMEM Ahn method

loq nlme

  • Please log in to reply
5 replies to this topic

#1 William R. Wolowich

William R. Wolowich

    Member

  • Members
  • PipPip
  • 27 posts

Posted 05 February 2016 - 04:14 PM

I have a C Vs t data set with several points below the GC/MS LOQ. These points are critical, as they appear in the lambda Z portion of the curve.

Ahn, J et al J PKPD August 2008, and then subsequently  Bergstand, M AAPS June 2009 recommend using a function in NONMEM called the F-Flag. For their M3 method, F-Flag is set to 0 (prediction) for samples above LOQ and set to 1 (a liklihood) for samples below LOQ.

How does one code the equivalent in Phoenix NLME, or is there a built in function similar to the M3 method?

Attached File  Likelihood based approaches to handling data below LOQ.pdf   421.21KB   1477 downloadsAttached File  samples below LOD in NLME.pdf   226KB   1271 downloads



#2 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 05 February 2016 - 04:34 PM

Dear Robert

We have built-in M3 method.

1:Click on BQL in the interface in the error model

2: In the data set, put the LOQ value in the CObs column when you have BQL and create a censor column called for example CObs BQL and put 1 when it is BQL.

An example is attached for you to review,.

Best

serge

Attached Files



#3 Gilles TUFFAL

Gilles TUFFAL

    Member

  • Members
  • PipPip
  • 14 posts

Posted 24 November 2016 - 08:34 AM

Dear Serge, 

 

I'm investigating the item of Conc below the LLOQ for NLME modeling.

I have imported your example and understood the settings but not the output.

In your example two values were set to the likely LLOQ and thus flagged as censored in the CObsBQL column. M3 is supposed to estimate likelihood of this values and thus to provide an IPRED/PRED value necessarily below the LLOQ.  Within outputs I don’t see any of this values. Just one line in residuals of the non-censored values. But the model accounts for likelihood of censored values I guess.

Sorry for my candidness. Which information can one expect from M3 use.

 

By the way, can M4 method be programmed in textual mode?

 

Thanks.

 

Gilles



#4 mittyright

mittyright

    Advanced Member

  • Members
  • PipPipPip
  • 98 posts

Posted 02 December 2016 - 04:39 PM

Hi Gilles,

 

I'm not so experienced as Serge

 

For M4 I would substitute observe() statement to the following one:

 

      LL(CObs, ((C > LLOQ) ? lnorm((CObs-C), sigma())- ln(1-phi(-1/sigma()))
                    : lphi((LLOQ-C), sigma())+ 0.5 * phi(-1/sigma()) - ln(1 - phi(-1/sigma()))))
 
Hope it helps,
Mittyright
PS please note I assumed additional error. For other type of residual error it is more complicated but doable. 

Edited by mittyright, 02 December 2016 - 04:40 PM.


#5 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 03 December 2016 - 09:51 AM

Dear Gilles

I am attaching a presentaiton that explains in general how to deal with this kind of problem.

The example assumes interval censoring which is more general.

I think you can easy go from my example to your specific case.

Best Regards

Serge

Attached Files



#6 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 03 December 2016 - 09:51 AM

I mean right censoring







Also tagged with one or more of these keywords: loq, nlme

2 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users


    Bing (2)