18 replies to this topic

[Zancong Shen]

I learned through the new phoenix NLME and found myself impressed by its functionalities. As I have reached the final model, I was puzzled on how to run a simulation with the final population PK parameters and covariates.

For example, how to simulate PK curves using parameters from final model in a larger population (eg., set N=500 subjects) with different covariates and stratifications, or simulate on new dose levels, or to steady states.

The Run options provide a simulation function but I don't see how useful it is.

Appreciate any suggestion from experienced modelers using NLME.

 

PKbeginner

[Simon Davis]

Hi PKBeginner,

   I'm not sure I entirely follow your question. When I have reached a final model in Phoenix I normally make a copy of that object first, (so I can leave my final run unchanged).

 

   In the copied object I then set the Run option to Simulate and then feed in my Final Parameter estimates from the final run to the simulation object. I can then change dosing regimen etc. in my copied object for different simulations.

 

   Does that help?

                                Simon.

[Ana Henry]

Phoenix NLME currently has some simulation functionality using the predictive check option.

You can freeze your final parameters estimates and fix the number of iterations to zero and check predictive check this will enable you to simulate with random etas and sigmas.

Phoenix will use the mapped to main “data” as a template for simulation i.e.: To change the number of N subjects modify your mapped data set to reflect your design change e.g:

• different N of subject

• different covariate distribution

• different dosing schedule: you can use the dosing sheet to easily incorporate dosing into your simulation you do not need to modify the main data each time. The ADDL and Steady state functionality will allow you to concisely specify multiple dose or steady state dosing.

 

Should you want to simulate more elaborate trial design Trial Simulator is the software of choice for its great flexibility. In the future Trial Simulator will be ported into phoenix to seamlessly simulate from NLME models with uncertainty and flexible design specification.

 

Ana Henry

Pharsight Corp.

[Zancong Shen]

Thanks both Simon and Ana for the useful suggestions. If I use the mapped dataset to simulate eg.400 subjects, I have to mannually generate a huge dataset. It would be flexible to just define N, time points, dosing regimen, covariate range and distribution to avoid generating these individual dataset for simulation, Does NLME provide this functionality or there are some tips to achieve this? In later version of NLME, will it incorporate Monte Carlo simulation functionality? Sorry if these are silly questions.

Thanks

Zan

[Ana Henry]

Dear Zan,

Currently one has to generate these individual datasets. Until Phoenix integrates Trial Simulator one can generate the template datasets using R or s-plus scripts relatively easily. This can be generated within these tools and then importing it to Phoenix or using Phoenix Connect.

 

Predictive check is a kind of monte carlo simulations and in later versions of Phoenix NLME and Phoenix TS more types of simulation and options will be supported.

 

Ana

[Zancong Shen]

Thanks, Ana. Look forward to the new features.

zan

[Zancong Shen]

Another puzzle from my simulation experience on NLME:

 

I was trying to simulate 20 individual concentration profiles using thetas ,etas and epsilon values from final model and keep covariates and doses same, just to see how variable the simulation is due to random/residual effects.

This is how I did:

Generated dataset listing ID (1 to 20), same covariates (wt, 70 kg, age 40), same dose (40 mg) and time points (1 to 24hr)

Chose same PK model as in final model. Allowed residual error Stdev value (eg. additive, 2.0) on the structure tab

On parameter--fixed effect tab, keep same thetas as in final model (freeze or unfreeze the values seem not changed the results)

Feed covariates values as in final model

In random effect tab, leave random effect (eg, 0.2, 0.4 for nV and nCl as in final model)

Run options--simulation, choose 300 datapoints, time 24hr and Y variable is C.

 

The simulation results always gave me identical simulated C for every subject thus I couldn't see any variations between these subjects.

 

What part was wrong on my simulation?

 

Thanks!

 

zan

[Ana Henry]

Dear Zan,

 

Please note that the simulation run option is not going to give you what you want for for population datasets (it can be used with individual modeling). For Pop datasets we suggest to use Pred. Check to do these simulations.

 

Ana

[raghava choudary]

Dear Beginner,

 

I was trying to simulate human Oral plasma concentration profiles from the predicted human pharmacokinetic parameters from rat and dog (human clearance and vdss predicted using allometric scaling from rat and dog).

 

while simulating oral profile there is no bioavailability included in the final parameters, so if i enter the values for clearance and volume of distribution, need to be corrected for bioavailability (i.e., CL/F and Vss/F).

 

please guide me,

 

Thanks with best regards,

 

Raghav,

[Zancong Shen]

Hi,

Sorry maybe the gurus here in the forum can give you better guidance on this. I am puzzled by this too!

 

OKbeginner

[Samer Mouksassi]

Hello,

 

Phoenix nlme can accommodate a bioavailability if you pass to a graphical or textual model.

If you give more details about what kind of scaling you are trying I might be able to help.

 

A dosepoint statement in your pml code or graphical model can have an argument bioavailibility.

 

 

Samer

[raghava choudary]

Dear Beginner,

 

I faced the similar problem while simulating the concentrations using simulation mode. The concentrations are similar in each subject (also found some times the concentrations are negative at the last time points).

 

I tried predictive check option and i could not find any change in the concentration (i.e no variability in the simulations).

 

Did u try the predictive check option for simulating the plasma concentration proofiles ???????/

 

This is how I simulated the Oral plasma concentration profile:

 

01) Generated dataset listing ID (1 to 50), same dose (10 mg) and time points (1 to 24hr)

 

02) Chose PK model - 1 compartment extravascular. Allowed residual error Stdev value (eg. additive, 1.0) on the structure tab

On parameter--fixed effect tab, entered estimated values (human predicted values from rat and dog),

In random effect tab, leave random effect (eg, 0.3, 0.3 for nV and nCl as in final model)

Run options-- predictive check, and in selected a table to simulate the data.

 

The final output is same concentrations in all the subjects.

 

Can u help me if u tried this / solved the problem ???

 

There were some posts on predictive check but i didnt really understood them.

 

Thanks with best regards,

 

Raghav.

[Zancong Shen]

Dear Raghav,

 

Sorry for the late response. I am rusted in PHX popPK modeling after a year. Haven't touched it for a while. Maybe other gurus can help?

 

Kind regards

 

PKbeginner

[serge guzy]

Dear Ragha

I am not sure what is the question but I looked at your previous e-mail where you used predictive checks and got same concentration values for each patient as if there were no variability across the population. Can you check the following

click on run options and I guess you choose predictive checks. now look on the right and you will see "freeze eta to po estimates". This check box should be unchecked. If it is checked, you will get all aptients with the same model parameters. Let me know if it resolve the problem.

best

serge

[Emily Colby]

Attached is a document describing simulations in NLME. [file name=Simulations_Final-20121010.docx size=1692945]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/Simulations_Final-20121010.docx[/file]

Attached Files

•  Simulations_Final-20121010.docx   1.61MB   818 downloads

Edited by Simon Davis, 22 October 2018 - 01:14 PM.

[Rosa Luo]

Wow Emily, thanks so much, this is very helpful!

 

Best,

Rosa

[Emily Colby]

Nathan,

 

If these options are greyed out it means you have a column mapped to "Sort" in Main. You need to have subject ID mapped to ID and nothing mapped to Sort if you want any of these options. It is tricky because when population is unchecked, you map subject -> Sort, but if population is checked, you map subject ->ID and not to Sort.

 

Emily

[Rosa Luo]

Hi Emily,

In my dataset, some subjects have PD observations, but missing PK concentrations which I need to simulation base on available PK data. Would it be possible only simulation the PK concentrations at the timepoints matches the PD observations from these subjects? I thought "Sim at Observations?" checkbox is for this, but it didn't seem to be. I mapped "time" to "time". Thanks!

Best,
Rosa