8 replies to this topic

[Davidq8]

Hi, 

I am trying to fit data to a model, but the model doesn't show a good prediction to a specific time point (5th plasma sample); I tried different structural and error models but still no improvements. I am attaching DV, IPRED vs. IVR to show the issue.

Can I get any help or suggestion on this issue?

Thanks 

 

Attached Files

•  DV IPED VS TIME.pdf   344.48KB   122 downloads

[Simon Davis]

David, i suspect this might be an artefact of your sampling schedule.  try requesting a table using somthing like seq(2,6,0.1) to give you richer profile around your Tmax and see how the predicted curve looks then.

 

really to look into this deeper we would need more GoF plots and ideally the original project.

 SImon.

[Davidq8]

David, i suspect this might be an artefact of your sampling schedule.  try requesting a table using somthing like seq(2,6,0.1) to give you richer profile around your Tmax and see how the predicted curve looks then.

 

really to look into this deeper we would need more GoF plots and ideally the original project.

 SImon.

 

Hi Simon, Thanks for your reply 

I am attaching the project file.

Thanks  

Attached Files

•  M1.phxproj   1.55MB   108 downloads

[Simon Davis]

please see the attached models with examples of table statements and also an example of overlaying fits. i think with the built in lib selections 3 com with lag and mult residual error is probably best but it's hard to be sure because we did not get CV% of params. I would consider if hitting CMax is as important as describing the rest of the curve well. generally it looks like a nice fit but maybe you could reduce complexity by fixing a parameter or assuming V2=v3? SImon.

Attached Files

•  M1_table.phxproj   5.4MB   107 downloads

Edited by Simon Davis, 21 June 2022 - 08:43 AM.

[Davidq8]

please see the attached models with examples of table statements and also an example of overlaying fits. i think with the built in lib selections 3 com with lag and mult residual error is probably best but it's hard to be sure because we did not get CV% of params. I would consider if hitting CMax is as important as describing the rest of the curve well. generally it looks like a nice fit but maybe you could reduce complexity by fixing a parameter or assuming V2=v3? SImon.

 

Thanks, Simon for your help, I really appreciate it.

I have a question regarding how to assume both V2 and V3 are having the same value ? how can I do that in the model?

Thanks   

[Simon Davis]

David - attached is a projec tshowin gyou a couple of ideas

 

I saw CL and CL2 had similar values so there is a text model that assumes they are the same.

 

alternatiely you could try freezing parameters e.g. Tlag that appear well defined and see if that helps, see the fixed effects tab of the final model.

 

in your case neither of those approaches seemed to help and I still did not get SE's

 

So I,

1) accepted final estimates from prior run.
2) set calc method to Fisher

3) and iterations to 1

 

this gave some SEs but we can see that they are too high and we have an indentifiability issue with 3 com model, so you may have to accept a 2 com although interestingly the poisson residual error might be OK even if not the best overall fit.

One last thin to try was the QRPEM algorithm and actually that looks OK.

 

I also notice that even with 3 coms there are are distinct bumps in the cwres vs Ivar plot so maybe some other structural approach should be considered?

If you had more than one dose level I would be tempted to look at saturable clearance - there is a sight suggestion of a convex curve. (you need at least 3 different doses)

also sampling more richly between 24 and 72 hours.

 

 Simon

 

 

 

Attached Thumbnails

• 
• 
• 

Attached Files

•  M1_table_freeze.phxproj   7.29MB   101 downloads

[Davidq8]

David - attached is a projec tshowin gyou a couple of ideas

 

I saw CL and CL2 had similar values so there is a text model that assumes they are the same.

 

alternatiely you could try freezing parameters e.g. Tlag that appear well defined and see if that helps, see the fixed effects tab of the final model.

 

in your case neither of those approaches seemed to help and I still did not get SE's

 

So I,

1) accepted final estimates from prior run.
2) set calc method to Fisher

3) and iterations to 1

 

this gave some SEs but we can see that they are too high and we have an indentifiability issue with 3 com model, so you may have to accept a 2 com although interestingly the poisson residual error might be OK even if not the best overall fit.

One last thin to try was the QRPEM algorithm and actually that looks OK.

 

I also notice that even with 3 coms there are are distinct bumps in the cwres vs Ivar plot so maybe some other structural approach should be considered?

If you had more than one dose level I would be tempted to look at saturable clearance - there is a sight suggestion of a convex curve. (you need at least 3 different doses)

also sampling more richly between 24 and 72 hours.

 

 Simon

Simon, thank you so much for your detailed explanation!!

Just one more question, you mentioned that other structural approach should be considered, since I have already tried two and three compartments models, are you referring to trying different absorption models or scenarios like transit compartments or weibull?

Edited by Davidq8, 25 June 2022 - 05:54 PM.

[Simon Davis]

yes or perhaps saturable clearance, entereo hepatci re-circulation.  Depending on what you know of your compound's metabolism

[Davidq8]

yes or perhaps saturable clearance, entereo hepatci re-circulation.  Depending on what you know of your compound's metabolism

 

Thanks, Simon for your explanation.

 

Best