NONMEM Users Network Archive

Hosted by Cognigen

RE: Two compartment model with fixed omega parameters

From: Rik Schoemaker <rik.schoemaker>
Date: Wed, 28 Jun 2017 10:17:54 +0000

Dear Waroonrat,

That's almost a philosophical question! One could argue that the data simpl=
y do not provide this information and so there is no way to estimate IIV fo=
r V2 and Q. One could also argue that perhaps FOCE-I does not provide the m=
ost accurate approximation to the likelihood possible, and that you could p=
erhaps get closer using SAEM or some other routine. I have always assumed t=
hat if FOCE-I cannot recover it, it is not there, but I have seen some thin=
gs lately that make me believe this may not be the case. Try SAEM and see i=
f that performs better :-).

Kind regards,


From: Waroonrat Sukarnjanaset [mailto:waroonrat
Sent: 28 June 2017 11:49
To: Rik Schoemaker <rik.schoemaker
Cc: nmusers
Subject: Re: [NMusers] Two compartment model with fixed omega parameters

Dear Martin and Rik,

Thank you so much for your helpful suggestions, I really appreciate it.

Could you please recommend me which estimation methods would be the possibl=
e methods to estimate IIV for V2 and Q?



From: Rik Schoemaker <rik.schoemaker>>
Sent: Tuesday, June 27, 2017 11:29 AM
To: Waroonrat Sukarnjanaset
Cc: nmusers
Subject: RE: [NMusers] Two compartment model with fixed omega parameters

Dear Waroonrat,

I fully support Martin's suggestions below, but to come back to your origin=
al question: the fact that IIVs are set to zero for V2 and Q does not mean =
the second compartment is 'gone'. If you would examine your model predictio=
ns, inclusion of the second compartment -even without IIV- would result in =
the characteristic bend in the elimination phase of your compound when view=
ed on the log-scale. It is quite often that NONMEM FOCE-I cannot estimate I=
IV for V2 and Q and fixing them to zero can result in perfect predictions o=
f your observed concentration profiles including the two-compartment behavi=
our. No-one would claim that these parameters are in fact the same for ever=
y single subject, just that the data cannot support making them different f=
or your subjects in this case.

Kind regards,

Rik Schoemaker, PhD

Occams Co÷peratie U.A.
Malandolaan 10
1187 HE Amstelveen
The Netherlands<>

+31 20 441 6410


From: owner-nmusers
Sent: 27 June 2017 11:30
To: Waroonrat Sukarnjanaset <waroonrat
Cc: nmusers
Subject: RE: [NMusers] Two compartment model with fixed omega parameters

Dear Waroonrat,

To know if a model "adequately describe the data" you need to study model d=
iagnostics. Read more for example here:

The AIC numbers in your case indicate that the two compartment model is a m=
uch better description of your data than the one compartment model (read on=
 AIC here: How=
ever, this is only a relative comparison and as pointed out before does not=
 say anything about whether any of the models "adequately describe the data=

All the best,

Martin Bergstrand, Ph.D.

Senior Consultant

Pharmetheus AB

From: owner-nmusers
Behalf Of Waroonrat Sukarnjanaset
Sent: Tuesday, June 27, 2017 10:52 AM
To: nmusers
Subject: [NMusers] Two compartment model with fixed omega parameters

Dear NMusers,

I have tried to find an appropriate base model.

I found that two compartment model with fixed Omega of V2 and Omega of Q =
= 0 (AIC 1860.17) provided smaller AIC than one compartment model (AIC 19=
21.83) did.

From these findings (no variability on V2 and Q), is it suggesting that one=
 cpt model could adequately describe the data?

I would truly appreciate it if you could give me some suggestions.

Kind regards,


(image/png attachment: image001.png)

Received on Wed Jun 28 2017 - 06:17:54 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: