8 replies to this topic

[wrubas]

Dear members,

 

I came a cross an issue in Phoenix. I select an error model (log additive) for PK. Then I duplicate the PK model and add the PD (all done in text mode).  While I keep the same rror model in text mode, I noticed in "Ind DVvs IPRED" that the correlation got worse and looked like an additive model, which I then confirmed in my PK only model.

Does this mean, in Phoenix only additive errors run even though in text you specify a different model? If so is there a way around?

 

Thank you,

 

Werner

[serge guzy]

Dear Werner

I tried to reproduce  a project similar to what you did and it is attached.

I do not see any problem.

Please take a look and tell me what is different in your model.

What I know is that only if you have one response that log-additive will transform your data in the log domain but not with 2 responses. May be it is related.

Best Regards

Serge

 

Attached Files

•  logadditivePKthenPDadd.phxproj   4.61MB   350 downloads

[serge guzy]

Sorry, something happened with the laptop.

Yes, did you fit data you got or di you also generate the data sequentially?

I can see a potential problem if you simulated sequentially because the PK log-additive will create simulated data in the log-domain while I think that when you add a second response, there is no log-transformation.

Then this can create confusion about what domain the data are created.

Anyway I simulated using multiplicative and additive for PD and then I fit with log-additive for PK  and additive for PD.

All went well.

best

Serge

[wrubas]

Sorry, something happened with the laptop.

Yes, did you fit data you got or di you also generate the data sequentially?

I can see a potential problem if you simulated sequentially because the PK log-additive will create simulated data in the log-domain while I think that when you add a second response, there is no log-transformation.

Then this can create confusion about what domain the data are created.

Anyway I simulated using multiplicative and additive for PD and then I fit with log-additive for PK  and additive for PD.

All went well.

best

Serge

Dear Serge,

 

Thank you for your response Here is the sequence how I did it:

 

1)Send PK data to PK model modify in txt (EHC) and using log additive error and obtain parameter

2) Send PKPD data to PKPD (Indirect inhib) and replace PK with above PK fix PK parameters see below

 

test(){   
 double(Switch)  # need to declare new vars in the sequence statement  
 deriv(A1 = (Ka * Agut) + (Ka * Aa)-(Cl * Cp) -(K1g * Cp * Vc))  
 deriv(Agut = Abile*Rate - (Agut * Ka))  
 deriv(Abile = (K1g * Cp * Vc)- Abile*Rate)  
 deriv(E = Kin * 1 / (1 + Imax * Cp / (Cp + IC50)) - Kout * E)  
 Cp = A1 / Vc  
 deriv (Aa = -Ka * Aa)  
 dosepoint(Aa)  
 sequence{E = Kin / Kout}  
 Biledump = 4.0  # start of recirculation modeled as a constant here based on obs data  
 error(CEps = 0.372)  
 observe(CObs = Cp * exp(CEps))  
 error(EEps = 1)  
 observe(EObs = E * exp(EEps))  
 Rate = Switch / Tau  
 sequence{  
  Switch=0; 
  sleep(Biledump); 
  Switch = 1; 
  sleep(Tau) 
  Switch = 0; 
 }  
 fixef(Vc = c(, 524.42, )) 
 fixef(Cl = c(, 376.5, )) 
 fixef(Ka = c(, 0.887109, )) 
 fixef(K1g = c(, 1.28579, )) 
 fixef(Tau = c(, 2.84951, )) 
 fixef(Kin = c(, 78, ))  
 fixef(Kout = c(, 4.2, ))  
 fixef(Imax = c(, 70, ))  
 fixef(IC50 = c(, 520, ))  
}   
 

Based on the inspection of IndDV vs IPRED Phoenix it is not running log additive but additive error for PK.

I appreciate your help.

 

Werner

[serge guzy]

Dear Werner

Can I get the project?

Now why do you Plot LnDV vs Ipred and not DV vs Ipred?

I believe that if you have 2 responses log_addiitve, the log_transformation does not show up in the output but only if you have only one response that is Log_additive.

Can you try just DV vs Ipred for the 2 responses and see if it works.

When you di only PK, then it was only one response and therefore DV and Ipred  are in the log Domain

 

Serge

[serge guzy]

Dear Werner

I continued to work on your project and decided to simulate using your code and see what is going on.

I first simulated your error model exactly as written.

These are the models showing up in the attached project in the workflow called "Workflow".

The fit was very good but I paid attention that you put an error for EEps of 1 which is approximately 100% cv which is huge.

Because of that, the estimated parameters for kin and kout were not as simulated one.

I then change EEps to 0.1 and simulated and fit using log additive for both Cobs and EObs.

Kin and Kout came up closer to the true simulated one.

Of course you have only one patient and therefore you cannot take advantage of the population approach which would give you with many patients a  better estimate of KIn and kout.

 

I do not see nay problem here.

The fit is as expected. The precision of the parameters depend on the error you put in the data and also if you have one patient or many patients.

You did not put any random effects and therefore I simulated only one patient.

Then to me all is good and I am not sure what problem you have.

The only way I can help you more is if you send me the data.

I hope all this was helpful for you.

best Regards

Serge

Attached Files

•  log_addiitve_pk_and_pd.phxproj   1.16MB   349 downloads

[serge guzy]

Thanks. I will work on that and will come back to you

Best

Serge

[Simon Davis]

Posting Serge's direct comments to help others, unfortunately can't post the project.

 

The data format had to be changed to make this working.

 

Here are the few key formatting rules you have to remember.

 

You have one column with the time that goes from 0 to 8, then again 0 to 8 , then again 0 to 8.

This is the typical case of cross over study where you have one patient going through 3 periods where he/she is getting the dose or 3 different patients.

I assume same patient as you did not put any random effects.

 

If you have indeed a cross over and full washout between periods, you need to tell the program the following:

1: do not sort the data by time. This is done by unchecking the sort option in run options

2: You need to tell the program that all the compartments must be reset to their original initial conditions every time you start a new period.

I added two columns, one called period and one called reset where I put a 4 when we start the period 2 and 3.

Then in input option I check the box called reset and put 4 to 4.

This means that when the program sees that 4 on the reset column, all the compartments are reset to their initial conditions(you Strat from fresh).

Since you do not have any random effects, I did it that way.

It means the program will try to find one set of parameters that explain all the data .

Of course you should not expect perfection but it is not so bad after all.

Let me know if all is clear and how I can assist you more.

Best Regards

Serge

[serge guzy]

Here we go the project attached as Werner OK with it.

Best Regards

Serge

sorry. Forgot the project

Serge

Attached Files

•  log_addiitve_pk_and_pd.phxproj   473.65KB   347 downloads