Jump to content


Photo

Model fitting with food effect data


  • Please log in to reply
7 replies to this topic

#1 joybaker

joybaker

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 30 November 2022 - 03:43 AM

Hi all,

I am trying to fit a model into a food study PK. This is a cross over study so each subject administered the drug under both fasted and fed state. The drug absorption delayed in fed state (see fig below). The model can fit both fed ONLY and fasted ONLY data very well. However, in the combined dataset, see IPRED vs DV, there are always several data points out of the line. These data are mainly fed state data at the beginning of the absoprtion phase and the delay is not well captured.

 

So I tried a few modification, set covariate in Ka and Tlag (m09), write code for both fasted and fed Ka and Tlag (m10, so I can give better initial value), and also since fed state seems to fit fine with one cmt, I set flag to make 2 cmt for fasted and 1 cmt for fed. However, the fitting is still not very idea with the combined data.

 

I would prefer to fit one model for the whole dataset. For fasted state, it is clearly a quick abosrption for 1 cmt and a slow for the 2nd cmt, while for fed state, it is a delay and then a slow 1 cmt. The elimination phase looks quite similar so I think it could fit into one model.

 

Do you have any suggestions?

 

For the figure, 1 for fasted and 2 for fed state

Attached Thumbnails

  • Rplot11.png
  • Rplot222.png

Attached Files


  • AlexKawn and WilliamJainy like this

#2 smouksassi1

smouksassi1

    Advanced Member

  • Members
  • PipPipPip
  • 231 posts
  • LocationMontreal

Posted 30 November 2022 - 09:18 AM

have you tried IOV ?

why is your DV at time 168 = 0 ?

 

do you have actual times instead of nominal times ?



#3 joybaker

joybaker

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 30 November 2022 - 02:14 PM

The cross over has a 7-day washout, so the BQL at the  dosing point is assumed to be zero.

Previously used NA, but want to avoid the figure below, change it to zero. Haven't used BQL/2 yet, but think that is not going to make the above fitting better.

 

I am not very familiar with the IOV setting, do you mean that I just simply set food effct as an occational covariate?

have you tried IOV ?

why is your DV at time 168 = 0 ?

 

do you have actual times instead of nominal times ?

 

Attached Thumbnails

  • 16696395923146.png


#4 joybaker

joybaker

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 01 December 2022 - 01:29 PM

I used reset data to redo the model.

I think something is going wrong, that Ka can be over 50? My previous fitting it is always around 1.

The time point is the best I can get. However I have return the dosing point (Time = 168) back to BQL/NA

have you tried IOV ?

why is your DV at time 168 = 0 ?

 

do you have actual times instead of nominal times ?

 

Attached Files


Edited by joybaker, 01 December 2022 - 01:31 PM.


#5 bwendt@certara.com

bwendt@certara.com

    Advanced Member

  • Administrators
  • 282 posts

Posted 09 December 2022 - 04:47 PM

It seems your reset was not set up properly: please note that a reset value of 0 means no reset should be performed.
I have recoded your dataset and replaced the 0s with 4s.
I have also changed your definition of Ka and Tlag parameters. These parameters are better expressed with occasion random effects. Overall, this shows an improvement in the fit:

model_comparer.png
dv_ipred.png

See new project file attached.

Bernd

Attached File  A0001.phxproj   3.09MB   150 downloads

#6 joybaker

joybaker

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 13 December 2022 - 03:37 AM

Hi Bernd,

 

Thank you for the suggestion. Is there a list of default numbers that means paticular rules apart from the zero you mentioned?

I read from the assistance library: 

"reset(reset_column_name=c(lowvalue, highvalue))
Example: reset(RESET=c(3, 4)) says that RESET is the data column signifying a reset­ting of subject time. If the value in the reset column is between three and four inclusive, time is allowed to be reset on that row. Also, all compartments in the model are reset to their initial val­ues."
So my understanding is that any values can be used as a reset as long as it is specified in the low and high value tab?
And why dose it have low and high values? To flag different types of reset?
 
 
For the IOV:
IOV is the interoccasion variability implemented with inter-individ­ual variability incorporated inside. From the equation, my understanding is that IOV shares the same typical value for the same subject in different occasion with different random effects. This would be suitable for cross over study say using reference and test formulation. But is that true that for very obvious food effect or DDI in cross over study, it may have PK parameter values centered around two different typical values in two occasions (so that rationally IOV is not a reasonable choice)? Say in my case, the fasted ka is below 1, and fed ka can be around 10. And fed states clearly have a tlag and fasted may not? Please do correct me if this understanding is wrong.
 
Anyway, the two typical values (Ka_fasted/Ka_fed) option is not very stable in my model as the initial values can change the fitting results quite significantly. The IOV seems to give more stability.
 
 
 
I used build-in model to check the IOV model text, and I see your code is slightly different. Any particular reasons to add the *exo(nKa) in the end?
 
tparm(Ka = tvKa * exp(nKa + nKax0*(FoodEffect==1) + nKax1*(FoodEffect==2))*exp(nKa))
tparm(Tlag = tvTlag * exp(nTlag + nTlagx0*(FoodEffect==1) + nTlagx1*(FoodEffect==2))*exp(nTlag))
 
build-in text
stparm(Ka = tvKa * exp(nKa + nKax0*(FoodEffect==1) + nKax1*(FoodEffect==2)))
 
Thank you
 

It seems your reset was not set up properly: please note that a reset value of 0 means no reset should be performed.
I have recoded your dataset and replaced the 0s with 4s.
I have also changed your definition of Ka and Tlag parameters. These parameters are better expressed with occasion random effects. Overall, this shows an improvement in the fit:

attachicon.gifmodel_comparer.png
attachicon.gifdv_ipred.png

See new project file attached.

Bernd

attachicon.gifA0001.phxproj



#7 bwendt@certara.com

bwendt@certara.com

    Advanced Member

  • Administrators
  • 282 posts

Posted 13 December 2022 - 09:51 AM

You can use several reset IDs for several difference reset events. However, in order to take effect they need to be different to 0. There are additional options to reset individual compartments using the dosepoint or the sequence statement.

 

Sorry, for the code, it is laziness on my side. I started with just the occasion random effect and after that wanted to try out an additional between-subject random effect. These are mathematically equivalent forms to write the exponential function.

 

Bernd



#8 joybaker

joybaker

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 24 July 2023 - 07:59 AM

What is the difference between regarding FE as a covariate or as an IOV?  I mean like in the report discussion part, how should it be explained to choose IOV over covariate (not from the OFV point of view). Will it also affect simulation and extrapolation?

 

In the above case the IOV fits better than covariate, while in other cases covariate works and we normally accept it. Is it because it has more freedom for random effect (multiple eta ie.nKa, nKax0 and nKax1 for each individuals, in covariate mode, only nKa for individuals) that it gives a better fit? 

 

 

Attached Thumbnails

  • 111.png





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users