**From: "Honghui Zhou" HZHOU@janwcc1.ssw.jnj.com
**

Dear NONMEM user friends

My name is Honghui Zhou. I am now working at Janssen Pharmaceutica, Dept. of Clinical Pharmacokinetics as a Scientist. I am trying to establish a two-compartment model with first order input and M-M output. The command stream is listed below. When I run it, the message I got is MINIMIZATION SUCCESSFUL, but the PRED values are 10,000 times less than the DV. I tried very hard, but the problem still exists. Maybe something wrong with the scale factor. Because I used differential equations in the command file, how can I use a appropriate scaling factor. Or there is other problem with my command file. I would appreciate if you could help me out.

Since we can not send out the attached files outside the company, I have to copy the command file & data file (only one subject) below. Sorry for the inconvenience. The dose is 200 mg, while the unit for DV is ng/mL. The drug was administered BID for 15 days, on Day 15, a whole plasma profile was captured.

Thank you very much in advance for your kind help. I am looking forward to your response.

Best regards,

Honghui Zhou, Ph.D.

Janssen Research Foundation

(609) 730-3011 (phone)

(609) 730-3044 (fax)

e-mail: hzhou@janus.jnj.com

$PROB ITR Nonlinear Kinetics Modeling $INPUT ID TIME AMT DV

$DATA 916.PRN

$SUBROUTINES ADVAN6 TRANS1 TOL=4

$MODEL COMP=(DEPOT,DEFDOS) COMP=(CENTRAL,DEFOBS) COMP=(PERIPH) $PK

VM=THETA(1)

KM=THETA(2)

K12=THETA(3)

K23=THETA(4)

K32=THETA(5)

S2=THETA(6)

$ERROR

Y=F+ERR(1)

$DES

C2=A(2)/S2

DADT(1)=-K12*A(1)

DADT(2)=-K12*A(1)-K23*A(2)+K32*A(3)-C2*VM/(KM+C2)

DADT(3)=K23*A(2)-K32*A(3)

$THETA

(0, 9, 1000)

(0, 330, 1000)

(0, 0.5, 10)

(0, 0.2, 10)

(0, 0.07, 10)

(0, 800, 5000)

$OMEGA 1.2

$EST SIG=5 MAXEVAL=9999 NOABORT PRINT=5 $COVARIANCE

$TABLE TIME DV VM KM K12 K23 K32

FILE=918a.out

$SCATTER PRED VS DV UNIT

1 0 200000 0

1 12 200000 0

1 24 200000 0

1 36 200000 0

1 48 200000 0

1 72 200000 0

1 96 200000 0

1 108 200000 0

1 120 200000 0

1 132 200000 0

1 144 200000 0

1 156 200000 0

1 168 200000 0

1 180 200000 0

1 192 200000 0

1 204 200000 0

1 216 200000 0

1 228 200000 0

1 240 200000 0

1 252 200000 0

1 264 200000 0

1 276 200000 0

1 288 200000 0

1 300 200000 0

1 312 200000 0

1 324 200000 0

1 336 200000 0

1 348 200000 0

1 360 200000 0

1 360.5 0 407

1 361 0 564

1 362 0 820

1 363 0 990

1 364 0 1028

1 366 0 822

1 368 0 864

1 372 0 557

1 376 0 535

1 384 0 419

1 408 0 268

1 432 0 214

******
**

**From: alison@c255.ucsf.EDU (ABoeckmann)
**

Honghui Zhou sent the following request for help. I haven't run the problem, but I can see an error in the differential equations:

DADT(1)=-K12*A(1)

DADT(2)=-K12*A(1)-K23*A(2)+K32*A(3)-C2*VM/(KM+C2)

DADT(3)=K23*A(2)-K32*A(3)

The sign in the second d.e. for K12 is clearly wrong. It should be:

DADT(2)= K12*A(1)-K23*A(2)+K32*A(3)-C2*VM/(KM+C2)

With no drug entering cmt. 2 from cmt. 1, it is not surprising that the levels are too low.

Honghui, please try with the corrected d.e. and see if this helps.

You did not include the $INPUT record, but from the looks of the data it must be something like

$INPUT ID TIME AMT DV

The dosing pattern is very regular. You might find that the run times are faster if you use a single steady state dose event record at TIME 360 with SS=1 and II=12 (on the assumption that, after approx. 30 doses, the subjects have reached steady state).

Alison Boeckmann

******
**

**From: alison@c255.ucsf.EDU (ABoeckmann)
**

Dear folks,

I've had second thoughts about the reply that I sent to Honghui Zhou.

It may well be that the sign of the first term in DADT(2) was correct in his original control stream but that he made a mistake when he copied it to the email message.

If that is the case, then I'd like to make this point about methodology in the use of NONMEM.

He says:

the message

I got is MINIMIZATION SUCCESSFUL, but the PRED values are 10,000 times less than the DV. I tried very hard, but the problem still exists. Maybe something wrong with the scale factor.

If there is any question that the model may be incorrect, DO NOT RUN THE ESTIMATION STEP!!

He should delete the $ESTIMATION record, and compare PRED vs. DV with the initial estimates of theta. Maybe the initial thetas are so poor that NONMEM cannot find its way to the appropriate miniumim. Or maybe with initial thetas that are known to be reasonable, it is clear that there is some scaling error. Or maybe the PRED vs DV scatter looks good with initial estimates, but during the estimation, the thetas are changed so radically as to lead NONMEM to a false minimum, in which case carefully chosen bounds for one or more theta may help NONMEM stay in a reasonable area of theta-space.

The point is: if the model is wrong, the final estimates of theta may be so distorted as to make it impossible to figure out why the predictions are so poor.

Alison

******
**

**From: alison@c255.ucsf.EDU (ABoeckmann)
**

This is my 3d. attempt at responding to Honghui Zhou's question. Maybe this time I'll get it right. Lew Sheiner spotted the scaling error.

Zhou says:

The dose is 200 mg, while the unit for DV is ng/mL.

In his model:

...

S2=THETA(6)

$DES

C2=A(2)/S2

He is using theta(6) as both the central volume in the computation of C2, and as the scaling term for the predictions. The unit of theta(6) is presumably L, whence C2 is mg/L or mcg/ml. But S2 has the same units, and yet his concentrations are ng/ml, or 1000 times the values that are compatible with his code.

He needs something like this:

...

V=THETA(6)

S2=V/1000

$DES

C2=A(2)/V

Now predictions are converted from mg/L to mcg/L or ng/mL by the 1000 in S2's denominator.