Jump to content


Photo

ESTIMATE DOSES BY SIMULATION


  • Please log in to reply
23 replies to this topic

#1 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 06 March 2013 - 07:19 PM

If we fit a 1 cpt model with first order elimination and first order absorption and volume to single dose plasma concentration time values after a given dose then one obtains the values for the fitted PK paramaters. Use of the fitted paramaters enables us to simulate plasma concentration values following multiple doses at time points.[

 

Now what I would like to do is the following. I have the following:

 

(a) the plasma concentration and time values after two doses {precise times known but the precise doses are not known} of the drug.

 

(B) the plasma concentrations and the 1 cpt model paramaters i.e. K01, K10 and V fitted paramater after one single dose of the drug (known amount).

 

I reason that I should be able to simulate the values of the two doses {Aa1 and Aa2} to fit the multiple dose plasma concentrations. The times of administration of Aa1 and Aa1 are known. Can this be done?

 

The inputs would be the values of the PK paramaters plus the dosing time of each of the two doses.

 

I think this can be done by a Phoenix model? I do not envisage using lag times to set the timing of the two doses. I enclose a photo plus a file. I need advice on how to do this since I do not understand the Phoenix system model very well. I envisage a worksheet with 0 and 4 hours {times of dose} and Aa1 and Aa2 being mapped to these times?

 

Also I enclose a file of my efforts. Advice is most appreciated,

 

 

Angus [file name=estimate_multiple_doses-20130306.phxproj size=365220]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/estimate_multiple_doses-20130306.phxproj[/file]


  • Thomasgaks likes this

#2 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 06 March 2013 - 07:55 PM

Dear Angus

Here is what I suggest you to do. Since you have an estimate of the 2 doses but not the exact value, you can just add an F for both dose input and fit the model to the multiple dose data and estimate these 2 F's. F in fact multiply the approximate dose input by its value. Here F would play the role of an estimator of the true dose rather than a bioavailibility factor.

The same can do in the simulaiton mdoe where you would select the values for these 2 F's.

Best Regards;

Serge



#3 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 01:26 AM

Thank you Serge: Yes I would have an estimate of F for each of the doses. Your suggestion sounds fine, since it seems to be very simple. As yet I have not worked with F in this graphical environment and do not know how to fit it into the picture. I am learning the Phoenix graphical approach.

 

Presumably one constructs the dose Table {using the estimates} and F is written into the graphical model and the program deteremines each F when fitting the multiple dose plasma concentrations.

 

I will look into including F,

 

Angus

 

301-869-1000



#4 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 03:02 PM

Serge: On thinking about this I recollect that I have done something like this in WinNonlin or Adapt.

 

Here is what I tried. I introduced the f1 and f2 as you suggest as additional paramaters. This makes K01, k10 and V , f1 and f2 as the list of paramaters. I put in estimatees for all the parmaters as INITIAL.{good estimates}

 

 Set the run mode to SIMPLE, which is a fitting step. The idea if to fit the doses and f to the plasma concentrations {arising fom the 2 doses}.

 

I put in estimates for the dose {and they are good since I know the dose}. The time in the mapping Table is set to ORANGE. the dose Table is set to Aa 1 10000000 ng at time 0 and Aa2 1000000 ng at 4 hours. The units were set to ng for dose and hr for time.

 

So I tried to run the program and I got a error rmessage "one or more columns required for time in dosing". I do not follow this since the Table I set up has a column for Time.

It seems I do not follow the structure of this program. I am thinking that this should be working.

 

So let me put up what I have now in a file.

 

Angus [file name=estimate_multiple_doses_3_7_2013.phxproj size=119182]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/estimate_multiple_doses_3_7_2013.phxproj[/file]



#5 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 03:52 PM

Furhthermore to above: I set up a similar multiple dose fitting in WinNonlin. I used the same plasma concentrations (after 2 doses) and time points. Did not include f1 or f2. . Used the fitted values of k01, k10 and V ( I have them) for the paramaters and the same doses ( 1mg each).

 

I completer the fitting in a few minutes. The 1 cpt model used the multiple doses at 0 and 4 hours to fit the plasma concentrations. For the fitting step the Time and concentration values were used by the model. So for Phoenix both time and concentration values should be used for simple.

 

I tried to repeat above work (with no f) in Phoenix model and I could not fit the 2 doses to the mutliple dose plasma concentrations. Same error message "1 or more columns required for time in Dosing".

 

So I could not do a fitting of the multiple doses to the plasma concentrations (after 2 doses at 0 and 4 hours).

 

It seems that the Phoenix approach is different and I do not understand what it is.

 

Angus



#6 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 04:29 PM

Dear Angus

I made some corrections with your model. please take a look and come back to me.

There were many errors. Now about the ke, initial estimates suggested a really larger ke than the one you put. The fit seems too good as somebody that would just simujlate data without error. You did not define an observation either.

Best Regards;

Serge [file name=ANGUS2SERGE_official_version.phxproj size=368153]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/ANGUS2SERGE_official_version.phxproj[/file]



#7 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 04:54 PM

Serge : the observations are in the data. I impoorted the Cobservations for the multiple dose along the the the corresponding time points. Perhaps this program does not work like winNonlin. I will check ou tthe values for Ke. I know waht it is because I have the data set and fitted it to get the values of K10, k01 and volume.

 

Angus



#8 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 05:01 PM

Dear Angus

We need to train you in Phienix and the best way is to go to a course we provide. You need to define a response in the model as you use the edit as graphical feature and in your mdoel there were no green box that define a continuous response. Now froma model perspective, if the true doses are the same, volume and F will be unidentifiable as the impact of the same factor increase or decrease from the estimated dose input can be corrected by just changing V. If I understadn correctly, you first esitmated Ka, Ke and V from a single dose with known dose input ans then shifted to multiple doses with unknown accurate dosing input. In htat case, to avoid identifiability issue, you must freeze (use fixed not estimable) the PK parameters other than f1 and f2. If you do so, then f1 and f2 will be identifiable.

Best Regards

Serge



#9 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 07 March 2013 - 05:04 PM

Angus, as Serge already indicated you had somewhere along the line lost this green box; Posted Image

that corresponds to this line in Serge's model code.

observe(CObs = C * (1 + CEps))

# (and also the error statement)
error(CEps = 0.1)

If you are fitting you would map observed concs to CObs.

The predicted or simulated Profile would be C.

Regarding training, you're in Maryland, right? You'll see future courses that are currently planned are here;

http://pharsight.com...ng_upcoming.php

but we may look to do something nearer you soon.

Simon.

PS try working through the NLME examples guide as the exercises like peripheral elimination and iv & oral simultaneous fit. Most can be run in non-popualtion mode too. Posted Image

#10 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 05:27 PM

Serge: I cerated the data set sett would be ewasy to work with. The parmamaters are fine.

 

I used these paramaters in Winnonlin to generate the plasma concentrations included in the worksheet. The plasma concentrations in the Phoenix worksheet I posted are following 2 doses and were imported from Excel into Phoenix. Thye should be removed?

 

    I think you are saying this is an error since they {the concentrations} should be included in the C obs like you show.

 

 I just checked them again for the single dose data 1mg or 1000000 ng {see below}.

 

for fitting

V_F 119903.39

K01 0.98180314

k10 0.97805983

 

 

If f1 and f2 are included they should solve as 1.0

 

angus



#11 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 06:08 PM

Dear Angus

When I simulate with your estimates, I am not getting at all the same profiles as the data set.

Best

Serge



#12 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 07:19 PM

Serge: That is intriguing. I remember that when I met you at AAPS you presentd a high level poster concerning POP PK with( I thought) your own program. I must be misleading you.

 

I feel there must be a misunderstandoing. Here is the data set I am working on in Phoenix WinNonlin model classic (1 cpt model with K10 and k10, V). I did a fitting of the single dose plasma concentrations (1 mg) and then I simulated 2 1 mg doses (0, 4 hours). These simulated plasma concentrations ( 1 mg bid) are the ones I was working with in Phoenix. I thouhgt it would be an easy set to work with. You can see for yourself the fit and the simulation.

 

I do understand your point very well about fixing the values of the PK parmaters to get the f.

 

 

I have been using classic WinNonlin for 18 years. Plesae let me know what you think.

 

 

Angus [file name=SERGE_MULTIPLE_DOSES.phxproj size=195645]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/SERGE_MULTIPLE_DOSES.phxproj[/file]



#13 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 07:23 PM

Simon: Thanks for the advice and information. I find the graphical approach attractive , but I am not a population fellow. I do like the field though as it continues to grow.

 

I will try some more.

 

Angus



#14 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 07 March 2013 - 08:05 PM

I took the project from Serge and added a Sim model.

I guess the questions that jump to mind involve wht happens if I try fitting the conc data from two doses you supplied - I get quite different estimates and high CV%

- probably it's too complicated a model and we should consider fixing some of the estimates to be the same; are you really dosing to two different places?

did your estimates come from single dose data fitting, do you have that data too?

Simon. [file name=ANGUSwithSim.phxproj size=722565]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/ANGUSwithSim.phxproj[/file] Posted Image

#15 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 08:59 PM

Dear Angus

I finally found the error in your model. It was a tricky one. Here it is.

You define the model parameter ke believing it is a microcosntant model parameter but in fact if you click on the model parameter box and look below the graphical area you will see that ke is defined as a clearance/volume parametrized parameter. You need to click on the other options and select microconstant.

OIf you do that and freeze all the PK parameters except f1 and f2, then the fit will elad to about 1 for both f1 and f2. I did it ans it worked perfectly.

The model is attached.

This error was difficult to track because I saw ke and did not pay attention it wa snot a ke but a clearance based parameter. When I fit the model to the data and saw ke being about 1000000, then I understood the problem.

Cl=V*ke and therefore V about 1000000 would lead to an artifical ke of 1000000 because it is a clearance parameter.

The project is attached.

Best Regards

Serge [file name=ANGUS2SERGE_official_version-20130307.phxproj size=700428]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/ANGUS2SERGE_official_version-20130307.phxproj[/file]



#16 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 09:03 PM

Simon: I am envisaging a capsule with equal amounts of the same drug in 2 different bead types. One bead( 1 mg) releases immediately at O time and the second bead( 1 mg) releases 4 hours later lower down in the intestine.

 

I envisage the K01 and K01, V as the same for each of the beads.

 

Let me enclose the WiNonlin file that shows the fitting and simulation 2 doses: that is where the paramaters come from.

 

I am trying to envisage a situation of how you can by simulation estimate the doses {when you have a good idea what they are} and you have the values for the multiple dose plasma concentrations plus the PK paramaters .

 

Now does this help?

 

 

Angus [file name=SERGE_MULTIPLE_DOSES-20130307.phxproj size=195645]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/SERGE_MULTIPLE_DOSES-20130307.phxproj[/file]



#17 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 09:06 PM

Dear Angus

I hope you got the corrected model I juts sent you. now you can just estimate f1 and f2 giving all other PK parameters fixed (freeze). You sue the esti nate of the dose input in your input file and f1 and f2 will give you the best fitted dose input (dose1*f1) and dose2*f2.

Best is to start with initial estimate for f1 and f2 of 1 as you have a good estimate of these doses. You must freeze all other PK aprameters to prevent non identifiability.

best

Serge



#18 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 09:19 PM

Sege: it wsa not really an error: I saw that CL appeared when setting up the model. I thought that I could change it to K10 or ke and the program would recognize it and correct over to ke!

 

I will look at your modification.

 

Thank you,

 

 

Angus



#19 serge guzy

serge guzy

    Advanced Member

  • Members
  • PipPipPip
  • 485 posts

Posted 07 March 2013 - 09:22 PM

Dear Angus

It is an error in the sense you gave as comaprison estimates of ke in your single dose model where ke has units of 1 per hour while the units of the ke you defined in the multiple dose mdoel had units of ml/hour (clearances). Therefore it caused the mismatch.

Now with the corrected model you wills ee that everything is fine.

best

Serge



#20 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 March 2013 - 09:52 PM

Thanks Serge: You see I wondered about CL when it appeared......did not see the way to change over . I assumed that the program would accept Ke or K10 by name.......

 

I like the opportunity to fix values of PK paramaters to get f1 and f2 during the fitting. So now you see that I set up an ideal example to explore this concept of obtaining dose by fitting when you know estimate of dose and fix values for PK parmaters and you have the multiple dose plasma concentrations and times of admininstration. Wonder how it would work in a non ideal situation when for example drug precipitates out of solution lower in the intestine

 

I do not think you can fix values of paramaters during fitting in classic WinNonlin. You can specify a tight range. I must look into this to see if it can be done.

 

 

So your original idea was very good.

 

Thanks vey much for your help;

 

Angus






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users