Jump to content


Photo

Individual contributions to overall -2LL with NLME

NLME -2LL IIV

  • Please log in to reply
4 replies to this topic

#1 Pablo Morentin Gutierrez

Pablo Morentin Gutierrez

    Newbie

  • Members
  • Pip
  • 2 posts

Posted 11 March 2015 - 10:49 AM

Hi all,

 

I am simultaneously fitting PK and PD data using a 2 comp PK oral model and an indirect response PD model with and Imax/IC50 relationship to the PK) The dataset is rich for each subject in both the PK and PD and comes from 100 subjects (~3500 data points in total). My model has 28 parameters that are estimated (including IIV, covariates, correlations and residual errors) and I’m using the QRPEM algorithm. The model runs successfully and fits the data well (after a few hours running…).

 

I’m now comparing between 2 rival models (the only difference is inter-individual variability in the IC50 parameter, assumed log-normal distributed). The model including the IIV has a significant lower -2LL than the one without it (drop of 83 units in the -2LL). The interesting observation is that IIV in the IC50 is estimated as 210% (with a 19% CV) with a nice distribution of ETAS and acceptable Shrinkage (0.35) Also, the population IC50 is estimated with a 38% CV.

 

On the rival model (without the IIV in IC50) the estimated population IC50 has similar value but better precision (7% CV). All other parameters have same values and comparable precisions between both models. Also VPC (1000 simulations) showed no difference between both models.

 

Is it possible in Phoenix to see the individual contribution of each subject to the overall -2LL (I think it might be possible in NONMEM?), I’d like to see if the drop in the -2LL when I include the IIV in the IC50 is mostly driven by 1-2 subjects?

 

Apologies for the long question for a potentially very short answer…?

 

Thanks in advance for your help

 

Pablo



#2 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 11 March 2015 - 11:50 AM

Dear Pablo

Although the individual contributions are calculated, they are not stored yet in none of the tables. I made that request and hope it will show up in upcoming versions. Can you try to copy and paste your model to the workflow and run few iterations with FOCE ELS. It may give you more accurate -2LL. QRPEM does not need to give you an accurate -2LL to get precise final estimates. -2LL is not a part of the QRPEM algorithm.

Best Regards

Serge



#3 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 11 March 2015 - 11:50 AM

Can you share the project



#4 Pablo Morentin Gutierrez

Pablo Morentin Gutierrez

    Newbie

  • Members
  • Pip
  • 2 posts

Posted 11 March 2015 - 12:35 PM

Can you share the project

 

Thanks Serge. I can share but need to anonimyze the dataset first



#5 smouksassi1

smouksassi1

    Advanced Member

  • Members
  • PipPipPip
  • 231 posts
  • LocationMontreal

Posted 17 March 2015 - 11:42 PM

you can access the individual likelihood contribution but you will need to go to the temporary folder where your nlme model was run::
check your model outputs: e.g. text output search for somethin like this :
C:\Users\smouksas\AppData\Local\Temp\Phoenix\DME_82~1>

usually a folder starting with DME at the following location where instead of smouksas you have your windows log in 
(C:\Users\yourwindowsusernametologin\AppData\Local\Temp\Phoenix)

now that you found where is the folder is :
LLbysub.dat is the file  your are looking for
below is what this file looks like for the classical theop data you will have individual LL as well as the total sum:
 

    1               0.000000
    2              17.644246
    3              17.711518
    4               9.635479
    5              14.435775
    6              21.991898
    7              12.575109
    8              11.057640
    9              12.531197
   10              13.841060
   11              12.894396
   12              10.781035
   13              13.456423
 totlaplace,sumllbysub= -168.555776  168.555776

 


  • Venkateswari likes this





Also tagged with one or more of these keywords: NLME, -2LL, IIV

0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users