7 replies to this topic

[LLLi]

Hi,

 

I have some questions about the simulation of multiple dose (MAD) from single dose (SAD) with NLME. I have been stuck for a long time and it is really frustrating. I searched the forum but still didn't get the answer. I hope someone here can help me. My situation and questions are as follows.

 

I have the results of several cohorts of SAD trials and the dose range is 0.1, 0.3, 1, 2, 4, 8 mg.

 

What I did to them is to send all the data to Phoenix Modeling.

---First I used individual built-in model (PK, clearance, extravascular, 1) to fit them and tried different residual error model and got the PK parameter for each subject (I chose mixed ratio).

---Then I copied and pasted the individual model and changed it to pop model by checking the population checkbox. So I got one set of PK parameters for all subjects.

---Next I checked the effect of covariates (Race, Age, Sex, Dose) with stepwise covariate search and found that Race-V, Race-Cl, Dose-Cl is the scenario to use. So I added these covarites to my pop model and rerun it. I got a new set of PK parameter for all subjects. The reason that I added Dose as a covariate is when I plot the AUC/Dose vs Dose for all subjects the plot is not horizontal at the high doses (4 and 8mg).

 

Now my questions are

---What should I do if I want to simulate the MAD for further trials? I think there are two options:

1) To use the individual PK parameter to simulate the MAD of each subject to get a range of multiple dose concentration; 

or 2) to use the pop PK parameter to simulate the MAD for each subject and get a range of MAD concentration. 

Which option is more reasonable? I think the option 2 is better. Or is there any other choice?

 

---I should do the simulation in individual mode or pop mode?

 

---What file should be mapped to the Main in Setup? My PK data or something else?  

 

---For the Dosing in Setup I just input the MAD regimen? Can I put different regimens at the same time? 

 

---I found a file named "simulation_official.phxproj" attached by serge guzy in this post (https://support.cert...er-to-simulate/). It looks like a individual simulation. I am confused with the dataset in Main. What does the time from 0.1 to 48 mean? The time range that we want to simulate?  (In the Ind Simulation plot, the time range is from 0 to 48). Why in run options the Max X is set to 24?

 

Sorry for this lengthy post and so many questions. I am a PK beginner and really really need your help. Any comments are appreciated. 

 

Thank you very much!

 

LLLi

Edited by LLLi, 27 July 2016 - 05:02 PM.

[smouksassi1]

Hi LLLi,

 

 

 

Hi,

Before any modeling you need to keep the decision you are trying to optimize with the model in mind.

We have the following: SAD trials with  dose range is 0.1, 0.3, 1, 2, 4, 8 mg.

are you interesting to know what would have happened in the individuals you already have should you have go with a Multiple dose scheme ? if yes then using these individual parameter will answer this question.

For this also individual modeling is fine.

When you fit a Population Model you are interested in the Population Typical tendency and its variance i.e the distribution of the population from which you only collected a sample of N subjects.

The pop model will give you population model parameters and variance ( omega) and individual parameters as well.

Of course optimizing the model need effort and you detected non linearity. Using dose might be fine as long as you don't try to extrapolate beyond observed dose limits. If you want to simulate for untested doses you need to have a more mechanistic model of the non linearity.

If you want to simulate upcoming trials then you want to simulate "new" subjects from the distribution you fitted ( Pop Model) that can be done using the simulation mode in pop pk model ( you cannot just use subjects at hand this answer only question about these subjects). Use the simulation mode table to output for the range of time of interest.
Ideally you should take into account Parameter Uncertainty ( standard errors what you estimate as population clearance is not the perfect truth there is an error on it) and Model Uncertainty (your current best model might just be the best because you did not test the true better model ) but these things can be skipped for your initial simulation exercise) 

 

---For the Dosing in Setup I just input the MAD regimen? Can I put different regimens at the same time? 

yes once you set up your parameters you can include any input of choice ( dose, distribution of covariate information etc.) you can simulate at steady state or the each dose so you see how you attain the steady state.

 

Samer

[LLLi]

Hi smouksassi,

 

Thank you for your reply.

 

Before any modeling you need to keep the decision you are trying to optimize with the model in mind.

We have the following: SAD trials with  dose range is 0.1, 0.3, 1, 2, 4, 8 mg.

are you interesting to know what would have happened in the individuals you already have should you have go with a Multiple dose scheme ? if yes then using these individual parameter will answer this question.

For this also individual modeling is fine.

 

My purpose is to use my current SAD data to simulate the MAD regimen in new subjects. So that means I need use pop model as you mentioned below?

When you fit a Population Model you are interested in the Population Typical tendency and its variance i.e the distribution of the population from which you only collected a sample of N subjects.

The pop model will give you population model parameters and variance ( omega) and individual parameters as well.

 

After I ran the pop model, I only found a set of PK parameters in theta. Can we get individual parameters from pop model?

 

Another confusing issue is omega. I always get 1 in the omega even I tried to fun the example in Phoenix NLME 1.3 User's Guide. Some problem with my software? 

 

Of course optimizing the model need effort and you detected non linearity. Using dose might be fine as long as you don't try to extrapolate beyond observed dose limits. If you want to simulate for untested doses you need to have a more mechanistic model of the non linearity.

 

In the Phoenix NLME 1.3 User's Guide (page 23), it is said that select the Closed form have some disadvantages and one is that the differential system being converted must be linear (no nonlinear kinetics). Dose this mean that if we deselect the Close form we can pool all data no matter they are linear PK or nonlinear PK?

 

If you want to simulate upcoming trials then you want to simulate "new" subjects from the distribution you fitted ( Pop Model) that can be done using the simulation mode in pop pk model ( you cannot just use subjects at hand this answer only question about these subjects). Use the simulation mode table to output for the range of time of interest.

 

That is exactly what I want to do!

 

When I use pop model to simulate what file should be mapped to the Main in Setup?

Ideally you should take into account Parameter Uncertainty ( standard errors what you estimate as population clearance is not the perfect truth there is an error on it) and Model Uncertainty (your current best model might just be the best because you did not test the true better model ) but these things can be skipped for your initial simulation exercise) 

 

My estimate of  dVdRACE2 and dCldRACE2 are negative. And most CV%s are more than 30%. I tried to use 2 compartment model but it didn't improve the result. What can I do?

 

---For the Dosing in Setup I just input the MAD regimen? Can I put different regimens at the same time? 

yes once you set up your parameters you can include any input of choice ( dose, distribution of covariate information etc.) you can simulate at steady state or the each dose so you see how you attain the steady state.

 

So in the Parameters I can use the pop parameter?

 

Is the distribution of covariate information a must in my regimen? Now I just want to simulate different dose regimens and the new subjects will be enrolled with the similar criterion. 

 

Can I input different MAD schedule like this:

 

regimen Time Dose

1   0   1

1  168   1

1    336    1

1    504    1

1   672  1

1   840  1

1   1008  1

1    1176  1

1   1344 1

1   1512 1

1   1680  1

1   1848  1

1   2016  1

2   0     1.7

2   168 1

2   336 1

2   504 1

2   672 1

2   840 1

2   1008 1

2   1176 1

2   1344 1

2   1512 1

2   1680 1

2   1848 1

2   2016 1

 

 

Thank you again for your help!

 

LLLi

Edited by LLLi, 27 July 2016 - 06:29 PM.

[smouksassi1]

 

Hi smouksassi,

 

Thank you for your reply.

 

Before any modeling you need to keep the decision you are trying to optimize with the model in mind.

We have the following: SAD trials with  dose range is 0.1, 0.3, 1, 2, 4, 8 mg.

are you interesting to know what would have happened in the individuals you already have should you have go with a Multiple dose scheme ? if yes then using these individual parameter will answer this question.

For this also individual modeling is fine.

 

My purpose is to use my current SAD data to simulate the MAD regimen in new subjects. So that means I need use pop model as you mentioned below?

When you fit a Population Model you are interested in the Population Typical tendency and its variance i.e the distribution of the population from which you only collected a sample of N subjects.

The pop model will give you population model parameters and variance ( omega) and individual parameters as well.

 

After I ran the pop model, I only found a set of PK parameters in theta. Can we get individual parameters from pop model?

 

Another confusing issue is omega. I always get 1 in the omega even I tried to fun the example in Phoenix NLME 1.3 User's Guide. Some problem with my software? 

 

Of course optimizing the model need effort and you detected non linearity. Using dose might be fine as long as you don't try to extrapolate beyond observed dose limits. If you want to simulate for untested doses you need to have a more mechanistic model of the non linearity.

 

In the Phoenix NLME 1.3 User's Guide (page 23), it is said that select the Closed form have some disadvantages and one is that the differential system being converted must be linear (no nonlinear kinetics). Dose this mean that if we deselect the Close form we can pool all data no matter they are linear PK or nonlinear PK?

 

If you want to simulate upcoming trials then you want to simulate "new" subjects from the distribution you fitted ( Pop Model) that can be done using the simulation mode in pop pk model ( you cannot just use subjects at hand this answer only question about these subjects). Use the simulation mode table to output for the range of time of interest.

 

That is exactly what I want to do!

 

When I use pop model to simulate what file should be mapped to the Main in Setup?

Ideally you should take into account Parameter Uncertainty ( standard errors what you estimate as population clearance is not the perfect truth there is an error on it) and Model Uncertainty (your current best model might just be the best because you did not test the true better model ) but these things can be skipped for your initial simulation exercise) 

 

My estimate of  dVdRACE2 and dCldRACE2 are negative. And most CV%s are more than 30%. I tried to use 2 compartment model but it didn't improve the result. What can I do?

 

---For the Dosing in Setup I just input the MAD regimen? Can I put different regimens at the same time? 

yes once you set up your parameters you can include any input of choice ( dose, distribution of covariate information etc.) you can simulate at steady state or the each dose so you see how you attain the steady state.

 

So in the Parameters I can use the pop parameter?

 

Is the distribution of covariate information a must in my regimen? Now I just want to simulate different dose regimens and the new subjects will be enrolled with the similar criterion. 

 

Can I input different MAD schedule like this:

 

regimen Time Dose

1   0   1

1  168   1

1    336    1

1    504    1

1   672  1

1   840  1

1   1008  1

1    1176  1

1   1344 1

1   1512 1

1   1680  1

1   1848  1

1   2016  1

2   0     1.7

2   168 1

2   336 1

2   504 1

2   672 1

2   840 1

2   1008 1

2   1176 1

2   1344 1

2   1512 1

2   1680 1

2   1848 1

2   2016 1

 

 

Thank you again for your help!

 

LLLi

 

Hi again,

if you can share the project I can try to help more

 

My purpose is to use my current SAD data to simulate the MAD regimen in new subjects. So that means I need use pop model as you mentioned below?
 

SM: Yes.

 

After I ran the pop model, I only found a set of PK parameters in theta. Can we get individual parameters from pop model?

SM: Yes ask for the table and don't forget to ask for the parameters. If you are using a Race categorical effect you should get an StrCovCat  sheet. This will have the individual parameters values as well.

 

Another confusing issue is omega. I always get 1 in the omega even I tried to run the example in Phoenix NLME 1.3 User's Guide. Some problem with my software? 

You might not be using the population engine or something not mapped correctly.

We need to see the project to help.

 

In the Phoenix NLME 1.3 User's Guide (page 23), it is said that select the Closed form have some disadvantages and one is that the differential system being converted must be linear (no nonlinear kinetics). Dose this mean that if we deselect the Close form we can pool all data no matter they are linear PK or nonlinear PK?

 

Hi closed form only exist for some of the differential equation models so these are not related. As I mentioned you can use DOSE as a covariate on CL but in reality it is not a dose effect it is a receptor being saturated due to high concentration or something else the alternative to using DOSE would be to use a Michaelis Menten model, TMDD, non linear binding etc.

if you select saturating then your closed form solution disappear the software take care of the rest. 

Pooling all data is a user decision ideally you want to fit all data together so you integrate all the knowledge. The challenge is to come up with a model and theory that fit ALL the data.

 

That is exactly what I want to do!

When I use pop model to simulate what file should be mapped to the Main in Setup?

Just a file with the dosing regimen you want you can list all doses but you can also use the ADDL feature which is handy if you want regular dosing intervals. If there is covariate in your model you have to chose a covariate value that you want to simulate for

ID TIME DOSE ADDL II      RACE

1     0         1         2      24     2

 

equivalent to :

ID TIME DOSE    RACE

1     0         1             2 

1     24       1             2

1     48       1            2

 

My estimate of  dVdRACE2 and dCldRACE2 are negative. And most CV%s are more than 30%. I tried to use 2 compartment model but it didn't improve the result. What can I do?

I need to see how you parametrized you model to be able to comment. how many subjects you have ?

 

So in the Parameters I can use the pop parameter?

Yes and then you will be randomly sampling subjects based on the omega you estimated but first we need to fix why your omega are 1

 

Is the distribution of covariate information a must in my regimen? Now I just want to simulate different dose regimens and the new subjects will be enrolled with the similar criterion. 

Can I input different MAD schedule like this:

 

see above and it is up to you if you have other than race covariate effects you need to consider it think whether you want to simulate what happens for a 70 kg RACE =2 subject or whether you want to simulate a realsitic distribution of covariate including correlated weight, race,sex ( best distribution you can use if you have tons of data is your own data !) 

 

ID TIME DOSE ADDL II      RACE

1     0         1        13      168     2

2     0         1        13      168     2

3     0         1        13      168     2

4     0         1        13      168     2

5     0         1.7       0         0     2

5    168      1        12      168     2

 

 

[LLLi]

Hi mouksassi,

if you can share the project I can try to help more

 

I will ask my supervisor to see if I can share the data. He is out of town. 

 

1) My purpose is to use my current SAD data to simulate the MAD regimen in new subjects. So that means I need use pop model as you mentioned below?

SM: Yes.

 

2) After I ran the pop model, I only found a set of PK parameters in theta. Can we get individual parameters from pop model?

SM: Yes ask for the table and don't forget to ask for the parameters. If you are using a Race categorical effect you should get an StrCovCat  sheet. This will have the individual parameters values as well.

 

A Race categorical effect means that when we add a covariate we set its type as categorical when we run a pop model?

In my pop model with race as a covariate, I did find individual parameter in StrCovCat but the parameters for each subject are the same.  I think that is because I used naive pooled engine and eta=0 (please see the attachment named “omega”).

 

3) Another confusing issue is omega. I always get 1 in the omega even I tried to run the example in Phoenix NLME 1.3 User's Guide. Some problem with my software? 

You might not be using the population engine or something not mapped correctly.

We need to see the project to help.

 

I used Naive pooled engine. In the NLME guide, it is said that "All ETAs are forced to zero, and no OMEGA parameters are computed - only THETA and SIGMA" when using this engine. I think that is the reason. Naïve pooled engine is the only method in my software. Do other methods need extra license? If I only have Naïve Pooled engie, dose this mean that I can not run pop model?

 

4) In the Phoenix NLME 1.3 User's Guide (page 23), it is said that select the Closed form have some disadvantages and one is that the differential system being converted must be linear (no nonlinear kinetics). Dose this mean that if we deselect the Close form we can pool all data no matter they are linear PK or nonlinear PK?

 

Hi closed form only exist for some of the differential equation models so these are not related. As I mentioned you can use DOSE as a covariate on CL but in reality it is not a dose effect it is a receptor being saturated due to high concentration or something else the alternative to using DOSE would be to use a Michaelis Menten model, TMDD, non linear binding etc.

if you select saturating then your closed form solution disappear the software take care of the rest. 

Pooling all data is a user decision ideally you want to fit all data together so you integrate all the knowledge. The challenge is to come up with a model and theory that fit ALL the data.

 

If the model can not fit all data well can I split the pooled data into 2 parts (one include the data of 0.3-2mg, and the other includes the data of 4 and 8 mg) and then analyze them seperately? Do you have any other idea?

 

5) That is exactly what I want to do!

When I use pop model to simulate what file should be mapped to the Main in Setup?

Just a file with the dosing regimen you want you can list all doses but you can also use the ADDL feature which is handy if you want regular dosing intervals. If there is covariate in your model you have to chose a covariate value that you want to simulate for

ID TIME DOSE ADDL II      RACE

1     0         1         2      24     2

 

equivalent to :

ID TIME DOSE    RACE

1     0         1             2 

1     24       1             2

1     48       1            2

 

6) My estimate of  dVdRACE2 and dCldRACE2 are negative. And most CV%s are more than 30%. I tried to use 2 compartment model but it didn't improve the result. What can I do?

I need to see how you parametrized you model to be able to comment. how many subjects you have ?

 

I totally have 36 subjects.

 

7) So in the Parameters I can use the pop parameter?

Yes and then you will be randomly sampling subjects based on the omega you estimated but first we need to fix why your omega are 1

 

I am a little confused. Do you mean I need to input some values in the Ramdom Effect when I run the individual or pop simulation? These values are the values of nV, nCl etc from the Omega?

 

8) Is the distribution of covariate information a must in my regimen? Now I just want to simulate different dose regimens and the new subjects will be enrolled with the similar criterion. 

Can I input different MAD schedule like this:

 

see above and it is up to you if you have other than race covariate effects you need to consider it think whether you want to simulate what happens for a 70 kg RACE =2 subject or whether you want to simulate a realsitic distribution of covariate including correlated weight, race,sex ( best distribution you can use if you have tons of data is your own data !) 

 

ID TIME DOSE ADDL II      RACE

1     0         1        13      168     2

2     0         1        13      168     2

3     0         1        13      168     2

4     0         1        13      168     2

5     0         1.7       0         0     2

5    168      1        12      168     2

For “simulate a realsitic distribution of covariate including correlated weight, race,sex”, I am confused. Could you please explain it with more detail?

 

Furthermore, the example listed above is very helpful and I did the simulation accordig to this. Below are the steps and the results.

The regimen format looks like this

ID TIME DOSE   Dose 

1     0         1           1000  

1     24       1           1000

1     48       1            1000

2     0         1.7       1700      

2    24       1            1000

2    48       1            1000

3    0         0.5           500  

3     24       0.5          500  

3     48       0.5     500

Please see the attachment “regimen” for more detail. I only list ID time and dose because I only want to see the MAD concentration in subjects with similar covariates.

1) First, I simulated the MAD with individual model.

Map the sort, Aa (Dose in ng), and Time in Mappings.

Use the pop paramters (I ran two pop models, one is without covarites and the other is with covarites. Which set of pop parameter should I use? Here I use the pop parameter from the model without covarites).

Ramdom Effects: I didn’t input anything because I don’t know what to use.

Run options: #=1000, Max X=2016, and Y=C

The Residual error is mixed ratio

Run the model

It works! But I don’t know whether the results are correct since I didn’t use the ramdom effects. Please see the attachment “Ind simulation”.

 

2) Then I smimualted the MAD with pop model.

Copy and paste the individual model.

Check pop.

Change to ID=regimen and keep the remains.

N Iter=0 (actually I don’t know why I need do this. I saw this at other people’s post)

Run options: Main: #=1, X axis=t, Pred.=None; Add sim table: Times=seq(0,2016,1), variables=C

Run the model

The same doubt: the results are correct since I didn’t use the ramdom effects? Please see the attachment “XY plot pop simulation #1 none pred”.

 

3) Base on the mode 2) I changed something in Main in Run options. I set Pred.=additive and checked the Pred. Variance. Corr. Checkbox and run the model. The result of 3 is different from the result of 2 (Please see the attachment “XY plot pop simulation additive pred”).

The pred. is only for preditive? For simulaiton, we don’t need pred.? So set pred.=None (model2)) is more reasonable?

 

I compared the plots of 1) , 2) and 3). They are different. Which result I should use?

 

I also fitted all the data with PK libaray model (here I used uniform as weighting). I summarized the individual PK parameters and use the mean PK parameters (the CV% is huge) to simulate the MAD with PK libaray model.

 

WinNonlin is so powerful and provides so many choice so I am lost J Which one is better?

 

Thank you for your time again!

 

LLLi

Attached Thumbnails

• 
• 
• 
• 

Attached Files

•  regimen.xls   22.5KB   690 downloads

Edited by LLLi, 29 July 2016 - 07:43 PM.

[smouksassi1]

Dear LLLi,

I am on vacation the next two weeks so I will let other users help you.

You probably don't have a licence for the POP PK engine and you can only run naive pooled or by subject analysis which don't give you a good estimate of the variance in the population. so for now you cannot estimate nor simulate random effects !

The new model system is much more powerful than the old winnonlin system which did not support nonlinear mixed effects.

Regards,

Samer

[LLLi]

Dear LLLi,

I am on vacation the next two weeks so I will let other users help you.

You probably don't have a licence for the POP PK engine and you can only run naive pooled or by subject analysis which don't give you a good estimate of the variance in the population. so for now you cannot estimate nor simulate random effects !

The new model system is much more powerful than the old winnonlin system which did not support nonlinear mixed effects.

Regards,

Samer

 

Hi Samer,

 

Thank you for your reply and have a nice vacation!

 

LLLi

[LLLi]

To other users:

 

To save your time I summarize my questions. Any input will be appreciated. Thank you very much!

 

1) For the situation in which a model and theory can not fit ALL the data, what can we do? Is there any reference that I can read?

 

2) Running population model with Naive pooled engine will only get a set of pop parameter. This is the method called "Naive pooled approach"? How to do the "two-stage approach" in NLME? Run the individual model in NLME and then combine individual parameter estimates to generate the mean parameter?  

 

3) What is the difference between PK library model and Phoenix NLME built-in individual PK model?

What is in my mind is 1) the NLME PK model has more choice of weighting; 2) PK library model assume linear PK which NLME can analyze both linear and nonlinear PK. 

 

4) My purpose is to simulate MAD concentration in new subjects (with similar weight, sex, race , age etc to the old subjects) from several cohorts of SAD data. Please let me know whether my methods of simulation in my mind is right. 

---Run Population model with fixed effect, random effect and covariates  (assume I have other engines for pop model) and I get a set of pop parameters as well as individual parameters.  

---Simulation: copy and paste the pop model above and accept all fixed and random effect and remove the covariates.

The regimen format is like this:

ID TIME DOSE   Dose 

1     0         1           1000  

1     24       1           1000

1     48       1            1000

2     0         1.7       1700      

2    24       1            1000

2    48       1            1000

3    0         0.5           500  

3     24       0.5          500  

3     48       0.5     500

(Please see the attachment “regimen” for more detail)

Parameters: use the pop parameters (I also got indiviual parameters. What parameters to use here?)

Random Effects: use the random effect value from the pop model

Run options: Sim./Pred; Main: #=1, X axis=t, Pred.=None; Add sim table: Times=seq(0,2016,1), variables=C

(If I set replicate # to 100, I can have 100 sets of concentration-time data. I summarize the mean concentration-time and plot the figure. That figure is the final result?)

 

 

Thank you very much for your time!

 

LLLi

Edited by LLLi, 01 August 2016 - 07:09 PM.