From: Sam Wai Johnn <scip8115@nus.edu.sg>
Subject: Fortran errors
Date: Wed, 2 Jun 1999 18:42:39 +0800

Dear NONMEM Users,

While assessing a structural model, I encountered this execution time error:


run-time error M6201 : MATH
*** : DOMAIN error

It appears as soon as FSUBS is started. The control stream is as follows:

$PROBLEM INDEX SET DATA
$INPUT ID DATE TIME AMT RATE PREP ADDL II DV SS CMT AGE WT

HT=DROP
AST AL TOTB
$DATA DATA10.1
$SUBROUTINES ADVAN2 TRANS2
$PK
TVCL=THETA(1)+THETA(6)*(AGE-2.2)**THETA(7)
CL=TVCL*(1+ETA(1))
TVVD=THETA(2)
V=TVVD*(1+ETA(2))
IF (PREP.EQ.0) THEN
TVF1=THETA(3)
ELSE
TVF1=1
ENDIF
F1=TVF1*(1+ETA(3))
KA=THETA(4)
TVF2=THETA(5)
F2=THETA(5)
K=CL/V
S2=V
$ERROR Y=F+ERR(1)
$THETA 2 (0,30) (0,.2) (1.77 FIXED) (1 FIXED) 1.5 .2
$OMEGA .25 .25 .25
$SIGMA .000016

I am using NONMEM IV level 2.1 on Windows NT with MS Powerstation FORTRAN V4.0. I have 14 subjects and their age range between 1.1 to 14 years old (median=2.2 years old). The INTEL constants in BLKDAT was used which are the

correct ones for IBM PC.

It was thought that the run-time error may be caused by negative or zero operand being raised to a negative or zero power. Thus, EXIT statements were

added to the control stream :

IF (THETA(6).LE.0) EXIT 1 100
IF (THETA(7).LE.0) EXIT 1 200
IF (WT.LE.0) EXIT 1 300

When this new control stream was run with the data of the first 3 subjects using NONMEM V, it ended with this message:

MONITORING OF SEARCH:
0PRED EXIT CODE = 1
0INDIVIDUAL NO. 3 ID=0.30000000E+01 (WITHIN-INDIVIDUAL) DATA REC NO.1
THETA=
2.00E+00 3.00E+01 2.00E-01 1.77E+00 1.00E+00 1.50E+00 2.00E-01

PK SUBROUTINE: ERROR IN COMPUTATION ATTEMPT TO COMPUTE BASE**POWER WITH BASE<0 AND POWER>0

A symbolic debugger was used and these values were obtained at the point in

PK routine where the error is detected:

age = 1.659999966621399
(dbx) p age-2.2
age-2.2 = -0.5400000333786013
(dbx) p theta(7)
theta(7) = 0.2000000029802322

This shows that this computation, (AGE-2.2)**THETA(7) is mathematically undefined when age < 2.2.

I was wondering if anyone has encountered a similar problem before when "centering the covariates at their median"?

Can anyone suggest a better way of using age and theta(7) in the model to circumvent this run-time error?

Any suggestions would be much appreciated. Thanks.

WJ Sam
(e-mail: scip8115@nus.edu.sg)

****

Date: Wed, 02 Jun 1999 14:04:54 -0700
From: Lewis Sheiner <lewis@c255.ucsf.edu>
X-Accept-Language: en
MIME-Version: 1.0
To: nmusers <nmusers@c255.ucsf.edu>
Subject: Re: Fortran errors
Content-Transfer-Encoding: 7bit

"Centering" refers to the scale on which avariables is considered. If you are going to raise a variable to a power, then it is more natural to consider it on the log scale; hence centering should be AGE/2.2, rather than age-2.2 LBS.