9 replies to this topic

[raghava choudary]

Dear Colleagues,

 

I have IV and PO data at 3 different doses and the exposures indicate non-linear increase in exposures.

 

While i separately modeled IV data with MM kinetics the fit was good but when i included oral data also, the model parameters were not estimated. Find the attached project

 

Can anybody suggest !

 

Regards,

 

Raghav.

Attached Files

•  M&S_IVPO_280916.phxproj   879.67KB   334 downloads

[serge guzy]

Dear Raghava

I would suggest first to shift the engine to QRPEM and stop early when you see no change in -2LL(should be around 1652).

QRPEM is a very good algorithm when you want to get good final estimates but does not really about -2LL converging perfectly(it should oscillate and it is good enough).

I am running FOCE like you did and I do not see any problem other than you should never ask for standard errors before you are sure you at done with the fit.

 

Then do the following

Copy your model to the workflow

Shift to QRPEM and run it without se request and wait until you see no change in -2LL. Then click on "stop early"

Copy that model to the workflow, ask for 1 iteration and for se's.

It should work.

If still problems, let me know and I will send you the project working.

 

Now an advise, in general F must be between 0 and 1. It is better to put instead of F ,  ilogit(flogit) where flogit is a parameter you define as normally distributed(sum option in parameter/structural tab) and not default lognormal. If you do that, you are 100% sure bioavailability will be between 0 and 1

Best Regards

Serge

[serge guzy]

Dear Raghava

I would suggest first to shift the engine to QRPEM and stop early when you see no change in -2LL(should be around 1652).

QRPEM is a very good algorithm when you want to get good final estimates but does not really about -2LL converging perfectly(it should oscillate and it is good enough).

I am running FOCE like you did and I do not see any problem other than you should never ask for standard errors before you are sure you at done with the fit.

 

Then do the following

Copy your model to the workflow

Shift to QRPEM and run it without se request and wait until you see no change in -2LL. Then click on "stop early"

Copy that model to the workflow, ask for 1 iteration and for se's.

It should work.

If still problems, let me know and I will send you the project working.

 

Now an advise, in general F must be between 0 and 1. It is better to put instead of F ,  ilogit(flogit) where flogit is a parameter you define as normally distributed(sum option in parameter/structural tab) and not default lognormal. If you do that, you are 100% sure bioavailability will be between 0 and 1

Best Regards

Serge

Please look at the models called IVPO_2Comp_MM_Multiplicative err_serge_qrpem for the fit and

se_IVPO_2Comp_MM_Multiplicative err_serge_qrpem for se's

Note that DV vs PRED suggests small bias(model underestimates) but not dramatic.

SE's are quite big for Vmax and KM which is common and huge for Ka which is a sign you do not have too much info about the absorption phase.

Serge

Attached Files

•  M&S_IVPO_280916.phxproj   2.76MB   375 downloads

[raghava choudary]

Please look at the models called IVPO_2Comp_MM_Multiplicative err_serge_qrpem for the fit and

se_IVPO_2Comp_MM_Multiplicative err_serge_qrpem for se's

Note that DV vs PRED suggests small bias(model underestimates) but not dramatic.

SE's are quite big for Vmax and KM which is common and huge for Ka which is a sign you do not have too much info about the absorption phase.

Serge

Dear Serge,

 

Thank you for the reply and you are always supportive.

 

I am working on the project with additional data and i will get back to you with the project.

 

with best regards,

 

Raghav

[raghava choudary]

Dear Raghava

I would suggest first to shift the engine to QRPEM and stop early when you see no change in -2LL(should be around 1652).

QRPEM is a very good algorithm when you want to get good final estimates but does not really about -2LL converging perfectly(it should oscillate and it is good enough).

I am running FOCE like you did and I do not see any problem other than you should never ask for standard errors before you are sure you at done with the fit.

 

Then do the following

Copy your model to the workflow

Shift to QRPEM and run it without se request and wait until you see no change in -2LL. Then click on "stop early"

Copy that model to the workflow, ask for 1 iteration and for se's.

It should work.

If still problems, let me know and I will send you the project working.

 

Now an advise, in general F must be between 0 and 1. It is better to put instead of F ,  ilogit(flogit) where flogit is a parameter you define as normally distributed(sum option in parameter/structural tab) and not default lognormal. If you do that, you are 100% sure bioavailability will be between 0 and 1

Best Regards

Serge

Dear Serge,

 

Thank you for the reply and as guided by you worked on the project.

 

Please look at step 6 in the project: 2comp_IVPO_MM_Kafreeze_QRPEM_SE

I did freeze Ka value and modeled the data and the fit was good (Please comment).

Now for the simulation, can i use the QRPEM engine ? the no. of iteration to be used ?

 

Please look at step 6 in the project: 2comp_IVPO_MM_Kafreeze_QRPEM_SE_ilogitF (Please comment)

 

 

Thanks in advance.

 

Raghav

Attached Files

•  M&S_IVPO_290916.phxproj   7.96MB   374 downloads

[serge guzy]

Dear Raghava

No problem to freeze Ka. It is so fast that you cannot estimate it precisely and the large value you sued is OK.

 

If you simulate, you need first to change the engine to last option(sim/pred), then you can add a sim table and put the times you want as output.

You then put instead of the original data set a similar data set but with only the dosing information and not observations.

once you shift to the simulation option, there is no iterations performed (if you have at least phoenix 4.0).

To be sure you can put one iteration but it should not affect anything.

You can use any engine because there is no invocation of any engine when performing simulation.

Do not use naïve pool as may be the simulation will not use random effects.

try by yourself and you et into trouble, let me know and I will be happy to help you.

best Regards

Serge

[serge guzy]

Step 6 of the last model is OK too. In this case it seems that using directly F as lognormal or ilogit(flogit) with flogit being normal does not make any difference.

I would always opt for ilogit(flogit) especially if the variance of these random effects are large.

Note that you can define an expression called bio

bio=ilogit(flogit) and in the add/table option, you ask for bio and you will get the bioavailability for each patient.

If you define

biomean=ilogit(tvflogit) and as for biomean in the add/table option, you will get the average bioavailability.

flogit is the bioavailability in the logit domain and therefore can be negative or positive

best'

Serge

[Simon Davis]

Posting on Raghava's behalf;

 

Dear Serge,

 

Thank you for the guidance,

 

I tried simulating the data in the attached project (Step -7) Simulation (Please comment)

 

The parameters obtained are also scaled to higher species and simulations were performed. I have few questions around this ( Step 8 in the attached project).

 

01. The scaled PK parameters were entered in fixed effects tab (parameters) and what value should be entered in the random effects (is 0.3 (30%) ok for all parameters) ? any other suggestion ?

 

02. While simulating in Step-8 included a worksheet in the data containing doses for both i.v and p.o because the final model contains doses for both i.v and Oral. After simulation when i go to the table that includes simulated structural parameters the subjects that are indexed for i.v dosing also show bioavailability and Ka parameter in the tables. Is anything wrong? Can we consider that all subjects are dosed both IV and Oral ? Or how to address this?

 

03. When i use the project containing ilogit function for bioavailability the simulated structural parameters have bioavailability that had negative values? 

 

04. when is use the project containing normal log function for bioavailability the simulated structural paramters contain bioavailability values greater than 1 sometimes but not negative values. What is the difference ??

 

Thanks in advance

 

Raghava

[serge guzy]

There is no attached project.

Serge

[raghava choudary]

There is no attached project.

Serge

Dear Serge,

 

I have sent the project to simon by email as i cant upload the project here.

 

can you share e-mail so that i can send it thru.

 

Regrads,

 

Raghav