Jump to content


Photo

Assistance with Visual Predictive Checks -Population PK

population pk pharmacokinetics modeling and simulation

  • Please log in to reply
8 replies to this topic

#1 csheme

csheme

    Advanced Member

  • Members
  • PipPipPip
  • 59 posts

Posted 06 February 2017 - 08:55 PM

Hello all,

 

I am having difficulty generating respective VPC plots for my final population PK model.

 

I would like to have a VPC with a solid line representing the median of the simulated concentrations, with dashed lines representing the 5th and 95th percentiles. Shaded areas should correspond to the 95% confidence intervals for the prediction lines.

 

In my attached model you will see a PK model for plasma, liver, and kidney, for some reason the VPC is not working for me. Sometimes I can get it to work but only for plasma, not tissues, but even then the shading does not come through.

 

Is there a fault in this feature, or something I am doing wrong, could you please assist?

 

Please see the link to my project and images below with the last 2 models for 2 separate compounds,

 

https://www.dropbox....ey.phxproj?dl=0

 

8e3ca2a235.png

And here for the first compound, you see the entire confidence interval shading is missing for the 95% quantile, as well as nothing coming up for tissues. For the second compound nothing at all is coming up either for plasma or even any of the tissues.

 

90250012d5.png

 

 

 

Thank you greatly!


Edited by csheme, 06 February 2017 - 11:11 PM.


#2 mittyright

mittyright

    Advanced Member

  • Members
  • PipPipPip
  • 98 posts

Posted 07 February 2017 - 07:17 AM

Hi Csheme!

 

regarding VPC for plasma: 

please look at your XY axes, you are using semilog scale. Now look at PredCheck_ObsQ_SimQCI table. As you can see 5% quantile for 5% predicted quantile is negative!

That's the answer. PHX couldn't plot the shaded area for 5% predicted quantile.

 

regarding VPC for liver and kidney:

PHX fails in binning using K-means here. You can put your own bins or turn off this option 

 

hope it helps,

Mittyright


Edited by mittyright, 07 February 2017 - 07:17 AM.


#3 csheme

csheme

    Advanced Member

  • Members
  • PipPipPip
  • 59 posts

Posted 07 February 2017 - 07:15 PM

Thank you for explaining these to me. I am not sure why these are giving negative values... Is there anyway around this? Using the default seems easier, but is there a binning method that works best in these cases?

 

I did turn binning off, but still see the negative values.

 

I am most interested in obtaining a clear VPC for plasma, not so much tissues (these may be challenging).

 

Could it be something not right with my model causing these?


Edited by csheme, 07 February 2017 - 07:23 PM.


#4 mittyright

mittyright

    Advanced Member

  • Members
  • PipPipPip
  • 98 posts

Posted 07 February 2017 - 10:07 PM

Hi Csheme,

 

there are 2 different issues. First is related to shaded areas for plasma. The second is related to VPC for tissues (binning). 

If you turn off binning you do not change the confidence intervals for plasma.

Math doesn't care what kind of data you are using for building the model, so in case of uncertainty CI can go outside of semilog scale and PHX cannot do anything with it. 

 

By the way I noticed that you are using same residual error CEps for different observations. Compiler does not like it.

You may need to fix it like this:

error(CEps = 0.479721)

observe(CPlasma = C * (1 + CEps))
error(CKEps = 0.1)
observe(CKidney = C3 * (1 + CKEps))
error(CLEps = 0.1)
observe(CLiver = C2 * (1 + CLEps))
 
It won't change your results significantly but would be more correct
 
BR,
Mittyright


#5 smouksassi1

smouksassi1

    Advanced Member

  • Members
  • PipPipPip
  • 169 posts
  • LocationMontreal

Posted 09 February 2017 - 06:38 PM

currently the binning cannot be specified differently for each type of observation. you can always use your raw simulated data and do whatever binning/predcorrection you need using Phoenix , R or other tools.

 

The negative concentration need to be investigated more



#6 mittyright

mittyright

    Advanced Member

  • Members
  • PipPipPip
  • 98 posts

Posted 09 February 2017 - 08:54 PM

Hi Csheme,

 

Samer is right, you may need to investigate the model and change at least the type of residual error. It is sufficiently large for proportional error, that causes the negative values for simulated observations in VPC

 

Mittyright


Edited by mittyright, 09 February 2017 - 08:54 PM.


#7 csheme

csheme

    Advanced Member

  • Members
  • PipPipPip
  • 59 posts

Posted 22 February 2017 - 01:55 AM

Any clue on what type of error structure could be used, additive, multiplicative, mixed?

 

These should be multiple plasma concentration time profiles capturing multiple absorption distribution, trough and terminal elimination samples (i.e. concentrations differing by over 1 order of magnitude).

 

Now my model is in custom graphical mode, I do not see a way to change the error structure here, and if I convert back to bulletin all changes will be lost. What is the easiest way to make these adjustments?

 

 

Thank you all greatly.


Edited by csheme, 22 February 2017 - 02:07 AM.


#8 mittyright

mittyright

    Advanced Member

  • Members
  • PipPipPip
  • 98 posts

Posted 22 February 2017 - 12:09 PM

Hi Csheme,

 

Any clue on what type of error structure could be used, additive, multiplicative, mixed?

Any of them, also do not forget log-additive! Please note that the type of residual error could differ for observation types. Welcome to the world of modeling!

You need to investigate diagnostic plots and VPC data to choose the appropriate one (it is not so easy and straightforward!)

 

Now my model is in custom graphical mode, I do not see a way to change the error structure here, and if I convert back to bulletin all changes will be lost. What is the easiest way to make these adjustments?

 

click on the green rectangle with observation (say CKidney)

In the structure tab you see the name of observation, name of error variable and its type. You need to change the name of error (CKEps for example), select the type of error you want to explore for this type of observations and the initial value.

 

BR,

Mittyright


Edited by mittyright, 22 February 2017 - 12:10 PM.


#9 csheme

csheme

    Advanced Member

  • Members
  • PipPipPip
  • 59 posts

Posted 23 February 2017 - 07:57 PM

Thank you for your reply, much appreciated, I will investigate these! 

 

All the best!







Also tagged with one or more of these keywords: population pk, pharmacokinetics, modeling and simulation

0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users