From: "Gordi, Toufigh" Toufigh.Gordi@cvt.com
Subject: [NMusers] Dose and concentration unit differences
Date: Thu, 22 Sep 2005 14:56:10 -0700


Dear all,

I am a little bit puzzled by an observation I have made when running a
2-comp PK model and would appreciate comments. I have a data file, where
concentrations are given in ng/ml (which is the same as mcg/L). The dose
was given as mcg. I use the log-normalized data. When I use the data as
is, i.e. the data file contains the dose as mcg and the concentrations
as ln(mcg/L) the model doesn't run and has problem. However, if I change
the units and use ng for dose and ln(ng/L) for concentrations, i.e.
simply multiply both by 1000, the model runs and gives quite acceptable
results. Could anybody explain this? I give a couple of lines of the
data file below: 

The data without any manipulations: 

#ID  DATE=DROP   TIME    AMT (mcg)   DV (ng/mL)   EVID
 4   5/15/2000   10:19    400            .         1
 4   5/15/2000   10:25     .            15.9       0
 4   5/15/2000   10:37     .            7.66       0
 4   5/15/2000   11:04     .            4.28       0
 4   5/15/2000   13:45     .            1.18       0
 

Dose in mcg, Conc. in ln (mcg/L). 

#ID  DATE=DROP   TIME    AMT (mcg)   DV (ng/mL)   EVID
 4   5/15/2000   10:19    400            .         1
 4   5/15/2000   10:25     .            2.77       0
 4   5/15/2000   10:37     .            2.04       0
 4   5/15/2000   11:04     .            1.45       0
 4   5/15/2000   13:45     .            0.17       0 


Dose in ng, Conc. in ln(ng/L)

#ID  DATE=DROP   TIME    AMT (mcg)   DV (ng/mL)   EVID
 4   5/15/2000   10:19    400000         .         1
 4   5/15/2000   10:25     .            9.67       0
 4   5/15/2000   10:37     .            8.94       0
 4   5/15/2000   11:04     .            8.36       0
 4   5/15/2000   13:45     .            7.07       0 


Toufigh Gordi
_______________________________________________________

From: Leonid Gibiansky leonidg@metrumrg.com
Subject: [NMusers] Dose and concentration unit differences
Date: Thu, 22 Sep 2005 23:33:08 -0400

Hi Toufigh,
Is it a linear model ?
What exactly is the problem with the model, what type of error NONMEM returns?
Have you used any cut-off for the concentration near zero (something like if(F < 0.1) F=0.1 ?
Have you tried to increase precision of the computations (SIGDIG=5)?
Have you tried to start failing model from the parameter values given by the converged model ?
Do you have additive component of the error model?
It would be easier to debug the error with the control stream. In general, there should be
no difference in solutions, but some minor detail of the model may be not-scalable (or initial
values are not-scalable) leading to the differences in computations.
Leonid

_______________________________________________________

From: Partha Nandy partha.nandy@bms.com
Subject: [NMusers] Dose and concentration unit differences
Date: Fri, 23 Sep 2005 13:50:19 -0400


Hi Toufigh,

Did you scale the volume in order to get the concentrations in appropriate units ? At least
from the data set, it does not appear that anything is wrong, however something in the control file is probably causing the problem.

-Partha 
_______________________________________________________

From: "Gordi, Toufigh" Toufigh.Gordi@cvt.com
Subject: [NMusers] Dose and concentration unit differences
Date: Wed, 28 Sep 2005 14:06:48 -0700

Hi Leonid,

Sorry for the delay in my response to your mail. I was out of office for
a few days and also been working on this problem a little bit. I finally
managed to make the model work by a simple change from ADVAN9 to ADVAN8,
with identical parameter estimates, as suggested by you.

The control stream:  

$INPUT ID DATE=DROP TIME AMT DV EVID AGE SEXF WT 
CLCR NMS
$DATA LNPK2.csv IGNORE=#
$SUBROUTINE ADVAN9 TRANS1 TOL=5   ; or ADVAN8

$MODEL
NCOMP= 2
COMP = (CENT DEFDOS DEFOBS)
COMP = (PRF1)

$PK
"FIRST
" COMMON /PRCOMG/ IDUM1,IDUM2,IMAX,IDUM4,IDUM5
" INTEGER IDUM1,IDUM2,IMAX,IDUM4,IDUM5
" IMAX=1000000000

        SLP=THETA(1)*EXP(ETA(1))
        CLNR=THETA(2)
        CL=CLNR+SLP*CLCR
        VC=THETA(3)*EXP(ETA(2))
        CLD2=THETA(4)
        V2=THETA(5)

        S1=VC

$DES
        DADT(1)=-A(1)*CL/VC-A(1)*CLD2/VC+A(2)*CLD2/V2
        DADT(2)=A(1)*CLD2/VC-A(2)*CLD2/V2

$ERROR
        IPRD = F
        IF(IPRD.LE.0) IPRD=0.0001
        IPRED=LOG(IPRD)
        W     = SQRT(THETA(6)**2+THETA(7)**2/IPRD**2)
        IRES  = DV-IPRED
        IWRES = IRES/W
        Y     = IPRED + W * EPS(1)

$THETA (0, 1)         ;1 SLP  
$THETA (0, 5)         ;2 CLNR
$THETA (0, 14)        ;3 VC
$THETA (0, 52)        ;4 CLD2
$THETA (0, 37)        ;5 V2
$THETA (0, 0.1)       ;6 ERROR
$THETA (0 FIX)        ;7 ERROR

$OMEGA 0.01           ;1 IIV SLP
$OMEGA 0.01           ;2 IIV VC

$SIGMA 1 FIX

$ESTIMATION NOABORT MAXEVAL=9999 PRINT=10 METHOD=1 INTERACTION
$COVARIANCE
$TABLE ID SLP VC CL                 NOPRINT ONEHEADER FILE=PATAB121
$TABLE ID TIME IPRED IWRES          NOPRINT ONEHEADER FILE=SDTAB121
$TABLE ID SEXF                      NOPRINT ONEHEADER FILE=CATAB121
$TABLE ID AGE WT CLCR               NOPRINT ONEHEADER FILE=COTAB121

Toufigh
_______________________________________________________