9 replies to this topic

[Charlie Brindley]

I've set up a simple oral two-compartment model (attached). Central compartment is plasma concentrations and peripheral compartment is tumour concentrations. I can't get the model to fit both data sets (same profile for both data sets)?

 

By the way, this model nicely fitted the data using some old software (TOPFIT).

 

Charlie [file name=BGN_tumour.phxproj size=361704]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/BGN_tumour.phxproj[/file]

Attached Files

•  BGN_tumour.phxproj   353.23KB   0 downloads

[serge guzy]

Dear Charlie

I looked at your model. First of all each error model requires a different name for the sigma (CEPS). I FIXED THAT.

Now, the model itself cannot work because some processes are missing to explain what we see in the data. I expanded your model into a kind of indirect response model with baseline E0 to be estimated.

The model is the last model and you can see very good fitting properties.

I would be very surprised to see that the same model you defined worked with TOPFIT.

Anyway, Phoenix language is fully flexible as you can see with this model.

Let me know if it helps you.

best Regards

Serge [file name=BGN_tumour_SERGE.phxproj size=847270]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/BGN_tumour_SERGE.phxproj[/file]

Attached Files

•  BGN_tumour_SERGE.phxproj   827.41KB   303 downloads

[Charlie Brindley]

Hi Serge,

 

Many thanks for this modelling.

 

I'll need to spend some time to get my head around an Emax model for this type of data!

 

Surely, we should be able to fit the data to a classical compartmental model? I would have considered that the tumour concentrations could develop by a delayed uptake from the central compartment and sequestration resulting in higher tumour exposure with longer tmax?

 

The data was indeed fitted nicely by TOPFIT notwithstanding poor precision of some parameter estimates. However, I notice that some parameters were not estimated with good precision in the Emax model?

 

I would really appreciate some help in developing a more classic model to account for the relationship between plasma and tumour concentrations, if at all possible.

 

Best regards,

 

Charlie

[serge guzy]

Dear Charlie

Here is the issue. You cannot have just a central and a peripheral compartment and link the peripheral to the tumor observation. This is combining apples to oranges and also make the model unit mismatch. What I did is to use the same model as you did (clearance parametrization which is better) but then define an expression with the tumor predicted being proportional to C2. This then make the tumor profile depending on C2 while drug level not depending on tumor. The first model you proposed in addition of the mismatch unit also made drug level depending on tumor and not only tumor depending on drug level.

the new model is attached and you can see nice fit and also better standard errors.

It is not surprising that the se's are not perfect but this is due to the poor quality in the data.

I hope this model better matches your expectations.

Best Regards

Serge [file name=BGN_tumour_SERGE-20140506.phxproj size=848174]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/BGN_tumour_SERGE-20140506.phxproj[/file]

Attached Files

•  BGN_tumour_SERGE-20140506.phxproj   828.29KB   0 downloads

[Charlie Brindley]

Dear Serge,

 

I'm much more comfortable with this model, thank you.

 

The rate constants derived from the Phoenix model are very similar to those estimated from TOPFIT using tumour concentrations directly applied to the peripheral compartment. However, TOPFIT has problems estimating the volumes of C1 and C2, so that may reflect the issues that you have raised?

 

Best regards,

 

Charlie

[serge guzy]

You are welcome.

yes definitively, using the previous model would assume that the constant I added to go from Drug to tumor is one and therefore if you change the units, the model fails. If now you change the units in either PK or tumor, the constant will make the adjustment and you will get same goodness of fit.

Note that I got about 25% error which is quite good. There is no bias in the fit when fitting simultaneously. This was not possible without making the change I made.

best

Serge

[Charlie Brindley]

Hi Serge,

 

Excellent, that makes sense. One last question: how do I obtain the predicted values so that I can reproduce the model fit to the profiles?

 

Thanks again,

 

Charlie

[serge guzy]

Dear Charlie

Look at the results output. There will be a table called residuals. IPRED is the individual predicted value and DV is the observed one.

Good luck;

best

Serge

[Charlie Brindley]

Thanks Serge,

 

I was looking for a more detailed output but this is fine.

 

Charlie

[Simon Davis]

Charlie, you can request a "table" of specific timepoints on the Run Options tab. The syntax follow S-plus conventions and the seq function

Seq(from=, to=,every)

It also accepts a vector of times concatenated with the c function e.g:
c(seq(0.5,2,0.01),seq(3,24,0.05)) or c(1,2,3,4,5)

Note that this Table is only needed when fitting, with simulation where the WinNonlin Classic input of from to and Npoints still applies

Simon

Attached Thumbnails

•