**From yin~y@glaxowellcome.com Thu Mar 20 06:17:23 1997
**

Dear NONMEMers,

I need some help on a lag time problem with NONMEM. Before adding all those ETAs and dose differences, I used a much simpler model. Here is my NMT file.

$PROB XXXX CONCENTRATION

$INPUT ID AMT TIME DV=CONC

$DATA CMA02_2.TXT

$SUBROUTINE ADVAN4

$PK

KA = THETA(1)

K = THETA(2)*EXP(ETA(1))

K23= THETA(3)

K32= THETA(4)

S2 = THETA(5)*EXP(ETA(2))

S3 = THETA(6)

ALAG1=THETA(7)

$ERROR

Y = F*EXP(EPS(1))

$THETA

(0, 2, 300) ;KA

(0, 0.5,10) ;K

(0, 0.1, 20) ;K23

(0, 0.3,20) ;K32

(0,20, 1000) ;S2

(0,10, 1000) ;S3

(0,0.5,30) ;ALAG1

$OMEGA 50, 50

$SIGMA 100

$EST MAXEVAL=8000 PRINT=10 SIGDIGITS=4 POSTHOC

$COV

$SCAT (RES WRES) VS TIME

$SCAT PRED VS DV UNIT

It gave me an error:

PROGRAM TERMINATED BY OBJ, ERROR IN CELS

WITH INDIVIDUAL 1 (IN INDIVIDUAL RECORD ORDERING)

INTRAINDIVIDUAL VARIANCE OF DATA FROM OBS RECORD 1 ESTIMATED TO BE 0

MESSAGE ISSUED FROM ESTIMATION STEP

AT INITIAL OBJ. FUNCTION EVALUATION

After deleting

ALAG1=THETA(7),

and it's initial value, I got some reasonable estimations for K and Ka, not for others, but it was a big improvement. I knew there was a lag time from WinNonlin, but I am not sure how to use NONMEM to estimate it. Someone advised me that I need to use some statements like `IF lag time ....' but he was not sure how exactly it was or it may not be needed anyway. Could anyone give me some help?

Thanks.

Yin

Here are data for the first two subjects:

1 100 0 0

1 0 0.25 0

1 0 0.5 1.66

1 0 1 3.46

1 0 1.5 3.61

1 0 2 2.87

1 0 3 2.12

1 0 4 1.47

1 0 5.1 1.21

1 0 6 2.07

1 0 8 0.51

1 0 12 0.46

1 0 16 0.2

1 0 24.17 0.11

1 400 672 0

1 0 672.25 0

1 0 672.5 1.77

1 0 673 36.64

1 0 673.5 28.96

1 0 674.08 18.48

1 0 675 15.25

1 0 676 11.66

1 0 677 9.74

1 0 678 6.43

1 0 680 5.27

1 0 682 1.85

1 0 684 2.19

1 0 688 1.12

1 0 696 0.29

2 100 0 0

2 0 0.25 0

2 0 0.5 0.9

2 0 1 2.36

2 0 1.5 2.22

2 0 2 2.05

2 0 3 1.53

2 0 4 1.28

2 0 5 1.13

2 0 6 0.64

2 0 8 0.44

2 0 12 0.17

2 0 16 0.07

2 0 24.22 0

2 400 672 0

2 0 672.25 0.15

2 0 672.5 8.77

2 0 673 15.15

2 0 673.5 12.68

2 0 674 10.62

2 0 675 10.35

2 0 676 6.74

2 0 677.17 6.29

2 0 678 4.09

2 0 680 2.62

2 0 682 3.52

2 0 684 1.13

2 0 688 0.54

2 0 696.03 0.37

2 800 1344 0

2 0 1344.25 0.1

2 0 1344.5 3.41

2 0 1345 14.92

2 0 1345.5 18.86

2 0 1346 18.09

2 0 1347 22.42

2 0 1348 21.05

2 0 1349 16.66

2 0 1350 11.79

2 0 1352 7.95

2 0 1354 4.74

2 0 1356 3.18

2 0 1360 1.06

2 0 1368 0.34

****

**From alison Thu Mar 20 09:14:24 1997
**

Yin Yin sent a message (attached at the bottom) asking for help

with lag time.

There are two obvious errors.

First, the initial estimate for ALAG1 is .5. But the first observation event is at Time .25 Hence, with the initial estimates of theta, there is no dose at time .25, and the prediction F is 0.

The proportional error model

Y = F*EXP(EPS(1))

cannot be used when the prediction F is zero for some observation.

NONMEM's error message is exactly as he said:

INTRAINDIVIDUAL VARIANCE OF DATA FROM OBS RECORD 1 ESTIMATED TO BE 0

A simple change is to use the additive error model

Y = F + EPS(1)

(or a combined additive / proportional model).

In the data, the DV for the observation at TIME .25 is 0, and for the observation at TIME .5 it is 1.66. This seems to imply that lag time is between .25 and .5, so one might consider bounding ALAG1 between these limits. But this should not be necessary with the additive error model.

Second, his model includes

S3 = THETA(6)

but the $INPUT record does not list the CMT data item. Hence, compartment 2 (central) is default for both dose and observations, and there are no observations from compartment 3. The scale parameters S2 and S3 are used only to convert compartment amounts to predictions (typically, concentrations) when there is an observation from the given compartment. Because there are no observations from compartment 3, S3 does not enter into the model for any observation event, and so THETA(6) cannot be estimated.

Finally, Yin yin asks:

"Someone advised me that I need to use some

statements like `IF lag time ....' but he was not sure how

exactly it was or it may not be needed anyway."

No such statements are needed with PREDPP.