From: Batul Parta - batulparta@yahoo.com Subject: [NMusers] Nonmem model Date: 12/17/2003 11:51 AM Dear all, I am having problems getting NM to estimate any parameters in the following situation. I have data on bactericidal effects of an antibiotic. The PK is fixed (one comp.). PD consists of three compartments: sensitive (comp. 2), non-sensitive (comp. 3), and non- replicating (comp. 4). PD data are log-transformed in the data file. My observations are the sum of comps 2 and 4. The nonmem codes are given at the end of this mail. My problem is that NM does not estimate anything but the two THETAs associated with the error terms. It converges all right but nothing else is changed from the initial estimates. In my data file, I have the usual stuff (date, time, amount) plus a CMT, which is 1 for the dose and 2 for the first (and only first) PD observation. Beside the model given below, I have tried one with no log-transformed data as well as one with a hypothetical compartment which is the sum of compartments 2 and 4. All have the same problem. Starting with different initial values doe not help either. Any suggestions? Best regards and thank you for your help, B. Parta _______________________________________________________ From: Leonid Gibiansky - lgibiansky@emmes.com Subject: Re: [NMusers] Nonmem model Date: 12/17/2003 12:33 PM Batul I think you need initial conditions. As written, A2=A3=A4=0 at time = 0, is this correct ? Are there any source terms (doses to compartments 2, 3, or 4)? In not, then zero is the unique solution of the system for A2, A3, A4. Discrepancy between zero and your observations goes into the error term. What is the intended meaning of: CMT, which is ... 2 for the first (and only first) PD observation? I think it will be ignored since you explicitly stated that VBAC=A(2)+A(4) IPRD=VBAC Leonid _______________________________________________________ From: Lewis Sheiner - lbs@lewisbsheiner.net Subject: Re: [NMusers] Nonmem model Date: 12/17/2003 1:59 PM I agree with Leonid: You need a 'dose' of 1 unit into CMT = 2 at time 0 so there are some bugs for the drug to kill. LBS. _______________________________________________________ From: Batul Parta - batulparta@yahoo.com Subject: Re: [NMusers] Nonmem model Date: 12/19/2003 10:04 AM Thank you for the suggestions. I entered the first observation as dose and the model converges, estimating parameters. However, I am uncertain if the procedure is correct. If the observations are given as log-transformed data in the data file, what does nonmem add up, when adding the contents of the 2 compartments? Obviously, one needs to add the non-transformed DV. Does nonmem read the data file, converts the log-transformed data in the data file to non-transformed, adds them, and then transfers back to the log scale? Should I try entering the data as observed (not log-transformed) in the data file or is there any other way to make nonmem do the correct adding? Once again, thank you for you help! B. Part _______________________________________________________ From: Leonid Gibiansky - lgibiansky@emmes.com Subject: Re: [NMusers] Nonmem model Date: 12/19/2003 10:31 AM NONMEM is not transforming the data. All the computations are done in the original scale. Log transformation enters the model only here: IPRED=LOG(IPRD) Therefore, the dose (the first PD observation in your case) should be entered in the original scale. All other PD observations should be log-transformed. Leonid _______________________________________________________ From: GIRARD PASCAL - PASCAL.GIRARD@adm.univ-lyon1.fr Subject: Re: [NMusers] Nonmem model Date: 12/20/2003 8:36 AM All this is a little bit confusing ... You don't show your $INPUT but I guess (1) you have only one DV which is the log of your observation and (2) you never observe A(2) and A(4) so (3) your prediction is correctly defined as log(A(2)+A(4)) according to the log-transform of your DV. NONMEM never does any transformation on the data, unless you use some date or time format or call the PASS( . ) subroutine, Bonnes fetes, Pascal Girard _______________________________________________________