From: Elassaiss - Schaap, J. <*jeroen.elassaiss*>

Date: Mon, 15 Feb 2010 08:25:37 +0100

Hi Markus,

This message occurs according to the introduction to NM6 guide: "This

may happen when e.g. the final gradient vector is too large". This means

that there is the possibility of some inaccurate covariance matrix, see

also my post in the 2008 discussion that you mentioned

(http://cognigencorp.com/nonmem/current/2008-November/1235.html). So if

in your case the error estimates are in the neigbourhood of expected it

should be no problem.

At the same time this may hint about where in your model some

instability is located. Which parameter is associated with a large final

gradient and why does it have such a small error?

Another option is to rescale your parameters (see

http://www.cognigencorp.com/nonmem/current/2008-January/0772.html) as I

notice a more than 1000-fold ratio between your thetas 7 and 8, even

more between theta8 and omega3. See the discussion in

http://www.page-meeting.org/pdf_assets/4964-Elassaiss-Schaap%20-%20Equat

ions%20variability%20reporting%20PK-PD%20-%20Final.pdf for additional

benefits of log-transforming thetas.

Best regards,

Jeroen

-----Original Message-----

From: owner-nmusers

On Behalf Of Leonid Gibiansky

Sent: Sunday, 14 February, 2010 14:52

To: markus joerger

Cc: NONMEM

Subject: Re: [NMusers] computational issues preventing COV

Hi Markus

Sometimes you just cannot force $COV to converge using defaults. If you

use NONMEM 7, try $COV UNCONDITIONAL option Also, nommem should explain

somehow what is wrong (R matrix is singular, parameter is on the

boundary, or something similar). If this is the case, then you can try

MATRIX=S

Could you provide more details of the output ?

Thanks

Leonid

--------------------------------------

Leonid Gibiansky, Ph.D.

President, QuantPharm LLC

web: www.quantpharm.com

e-mail: LGibiansky at quantpharm.com

tel: (301) 767 5566

markus joerger wrote:

*> dear NONMEM community,
*

*>
*

*> The following MESSAGE:
*

*>
*

*> MINIMIZATION SUCCESSFUL
*

*> HOWEVER, PROBLEMS OCCURRED WITH THE MINIMIZATION.
*

*> REGARD THE RESULTS OF THE ESTIMATION STEP CAREFULLY, AND ACCEPT THEM
*

*> ONLY AFTER CHECKING THAT THE COVARIANCE STEP PRODUCES REASONABLE
*

OUTPUT.

*>
*

*> ...has already been discussed in 2008, and seems to be a computational
*

*> issue preventing successful COV (Prof. Nick Holford). I encounter the
*

*> same problem with an i.v. Gemcitabine model. COV is successful with
*

*> GEM and its metabolite dFdU. However, if I add intracellular
*

*> triphosphates (dFdCTP), minimisation is just fine, but no COV. This
*

*> has persisted after attempts to simplify model or computation (e.g.
*

*> fixing ETAs, using just one EPS, introducing simulated PK-curves
*

*> (based on PK-estimates) back into the model). Mass-transport to
*

*> triphosphate-CMPT is intentionally fixed very low (low
*

*> concentrations), and volume-of-distribution of intracellular
*

triphosphates scaled to V1.

*>
*

*> Thanks for support,
*

*>
*

*> kind regards, Markus
*

*>
*

*>
*

*> --
*

*> Markus Joerger MD PhD
*

*> Department of Medical Oncology
*

*> Rorschacherstr. 95
*

*> 9007 St. Gallen
*

*> markus.joerger *

*> markus.joerger *

*> phone: +41-765591070
*

*> fax: +41-714946368
*

*>
*

*>
*

*> CONTROL STREAM:
*

*> ------------------------------
*

*> ----------------------------------------------------------------------
*

*> -----------------------
*

*>
*

*> $DATA MO5_gem33.CSV IGNORE=C
*

*> $SUBROUTINES ADVAN5
*

*> $MODEL COMP=(GEM) COMP=(P1) COMP=(dFdU) COMP(dFdCTP)
*

*>
*

*> $PK
*

*>
*

*> TVCL1=THETA(1)
*

*> CL1=TVCL1*EXP(ETA(1))
*

*>
*

*> TVV1=THETA(2)
*

*> V1=TVV1*EXP(ETA(2))
*

*>
*

*> TVQ1=THETA(3)
*

*> Q1=TVQ1*EXP(ETA(3))
*

*>
*

*> TVV2=THETA(4)
*

*> V2=TVV2*EXP(ETA(4))
*

*>
*

*> TVCL2=THETA(5)
*

*> CL2=TVCL2*EXP(ETA(5))
*

*>
*

*> TVV3=THETA(6)
*

*> V3=TVV3*EXP(ETA(6))
*

*>
*

*> TVK40=THETA(7)
*

*> K40=TVK40*EXP(ETA(7))
*

*>
*

*> K12=Q1/V1
*

*> K21=Q1/V2
*

*> K13=CL1/V1
*

*> K30=CL2/V3
*

*>
*

*> RF=THETA(8)
*

*>
*

*> K14=0.000001*K13
*

*> V4=K14*RF*V1/K13
*

*>
*

*> S1=V1/1000
*

*> S3=V3/1000
*

*> S4=V4/1000
*

*>
*

*> $ERROR CALLFL=0
*

*> IPRE=-3
*

*> FLG1=0
*

*> FLG3=0
*

*> FLG4=0
*

*> IF(CMT.EQ.1)FLG1=1
*

*> IF(CMT.EQ.3)FLG3=1
*

*> IF(CMT.EQ.4)FLG4=1
*

*> IF(F.GT.0) IPRE=LOG(F)
*

*> Y=LOG(F)+EPS(1)*FLG1+EPS(2)*FLG3+EPS(3)*FLG4
*

*> W=LOG(F)
*

*> IRES=DV-IPRE
*

*> IWRES=IRES/W
*

*>
*

*> $THETA
*

*> (0,163) ;CL-gemcitabine.1
*

*> (0,11.9) ;V1.2
*

*> (0,95.1) ;Q1.3
*

*> (0,20) ;V2.4
*

*> (0,14.1) ;CL-dFdU.5
*

*> (0,28) ;V3.6
*

*> (0,0.21) ;K40.7
*

*> (0,646) ;RF.8
*

*>
*

*> $OMEGA
*

*> 0.37 ;CL1.1
*

*> 0.23 ;V1.2
*

*> 0.0094 ;Q1.3
*

*> 0.036 ;V2.4
*

*> 0.14 ;CL2.5
*

*> 0.14 ;V3.6
*

*> 0.27 ;K40.9.7
*

*>
*

*> $SIGMA
*

*> 0.07 ;IIV-1
*

*> 0.0355 ;IIV-3
*

*> 0.757 ;IIV-4
*

*>
*

*> $ESTIMATION PRINT=5 MAXEVAL=3000 METHOD=1
*

*> NOABORT POSTHOC SIGDIG=3 MSFO=MSF004BBAB
*

This message and any attachments are solely for the intended recipient. =

If you are not the intended recipient, disclosure, copying, use or =

distribution of the information included in this message is prohibited =

--- Please immediately and permanently delete.

Received on Mon Feb 15 2010 - 02:25:37 EST

Date: Mon, 15 Feb 2010 08:25:37 +0100

Hi Markus,

This message occurs according to the introduction to NM6 guide: "This

may happen when e.g. the final gradient vector is too large". This means

that there is the possibility of some inaccurate covariance matrix, see

also my post in the 2008 discussion that you mentioned

(http://cognigencorp.com/nonmem/current/2008-November/1235.html). So if

in your case the error estimates are in the neigbourhood of expected it

should be no problem.

At the same time this may hint about where in your model some

instability is located. Which parameter is associated with a large final

gradient and why does it have such a small error?

Another option is to rescale your parameters (see

http://www.cognigencorp.com/nonmem/current/2008-January/0772.html) as I

notice a more than 1000-fold ratio between your thetas 7 and 8, even

more between theta8 and omega3. See the discussion in

http://www.page-meeting.org/pdf_assets/4964-Elassaiss-Schaap%20-%20Equat

ions%20variability%20reporting%20PK-PD%20-%20Final.pdf for additional

benefits of log-transforming thetas.

Best regards,

Jeroen

-----Original Message-----

From: owner-nmusers

On Behalf Of Leonid Gibiansky

Sent: Sunday, 14 February, 2010 14:52

To: markus joerger

Cc: NONMEM

Subject: Re: [NMusers] computational issues preventing COV

Hi Markus

Sometimes you just cannot force $COV to converge using defaults. If you

use NONMEM 7, try $COV UNCONDITIONAL option Also, nommem should explain

somehow what is wrong (R matrix is singular, parameter is on the

boundary, or something similar). If this is the case, then you can try

MATRIX=S

Could you provide more details of the output ?

Thanks

Leonid

--------------------------------------

Leonid Gibiansky, Ph.D.

President, QuantPharm LLC

web: www.quantpharm.com

e-mail: LGibiansky at quantpharm.com

tel: (301) 767 5566

markus joerger wrote:

OUTPUT.

triphosphates scaled to V1.

This message and any attachments are solely for the intended recipient. =

If you are not the intended recipient, disclosure, copying, use or =

distribution of the information included in this message is prohibited =

--- Please immediately and permanently delete.

Received on Mon Feb 15 2010 - 02:25:37 EST