From: Nick Holford <*n.holford*>

Date: Fri, 22 Aug 2008 11:00:55 +1200

Hong-Guang,

It seems you are already aware of the publications that have

demonstrated that NONMEM's warnings and failure to run the $COV step are

not of much use for model evaluation. So please stop worrying about them

and focus on more important things e.g. is your data correct (as Mark

Sale suggested).

You wrote:

"If this is the case at any time or even at most time, the NONMEM run

would not seem to be too tough."

I dont really understand what you mean. If you mean that using NONMEM is

much easier if we ignore these termination warnings then I agree with

you. Pay attention to your data and the model predictions. Dont pay

attention to these warnings that do not have any useful relationship to

the purpose of modelling.

Nick

Hong-Guang Xie wrote:

*> Hi Nick:
*

*>
*

*> Thank you for your quick response to my inquiry email.
*

*>
*

*> In my control stream (see above), the initial values for the "fixed
*

*> effects" parameters (such as f13, f15, and f16) were obtained from
*

*> each terminated run after typing and running "nmctl runname" in order
*

*> to fix a more important problem --"minimization terminated". In
*

*> addition, I used to try the *constraints* as you suggested, and
*

*> finally the minimization was still "terminated" but not "successful"
*

*> (plus, COV: NONE). If the final parameters returned from each
*

*> terminated run were used as initial values used in the $THETA section
*

*> (as seen in my last email), both "minimization terminated" and high F1
*

*> values came together. It there a good solution to the two problems at
*

*> the same time?
*

*>
*

*> Your saying -- "you can ignore the rounding error and additional
*

*> problem warnings. They dont mean anything useful."-- means that we
*

*> do NOT need to care about them too much, even if minimization is
*

*> "terminated" and covariance step "aborted" or "NONE", right? My
*

*> understanding was from your abstract
*

*> (http://www.page-meeting.org/?abstract=992) and another similar
*

*> abstract (CPT, 2005; 77:p2) by Gastonguay MR et al. If this is the
*

*> case at any time or even at most time, the NONMEM run would not seem
*

*> to be too tough.
*

*>
*

*> I will follow your suggestions to check if the model fits the data
*

*> well later.
*

*>
*

*> Thank you,
*

*>
*

*> Hong-Guang
*

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

*> On 8/21/08, *Nick Holford* <n.holford *

*> <mailto:n.holford *

*>
*

*> Hong-Guan,
*

*>
*

*> You can ignore the rounding error and additional problem warnings.
*

*> They dont mean anything useful. You should look at your parameter
*

*> values to see if they are sensible (clearly they are not for 2 of
*

*> the bioavailibility fractions) and also use things like a VPC to
*

*> decide if the model is fitting the data (see
*

*> http://www.page-meeting.org/pdf_assets/8694-Karlsson_Holford_VPC_Tutorial_hires.pdf).
*

*>
*

*> To avoid unrealistic estimates for bioavailability you can try
*

*> setting initial estimates for bioavailability fractions to
*

*> something sensible and use constraints e.g. instead of this
*

*> (0,4.93) ;f13
*

*> write this
*

*> (0,.7,1) ;f13
*

*>
*

*> Nick
*

*>
*

*> Hong-Guang Xie wrote:
*

*>
*

*> Dear NONMEM Users,
*

*> Recently I combined six popPK studies (STDY) together as a
*

*> whole and tried to re-estimate the pop PK parameters of that
*

*> drug in the population with a wide range of the age and body
*

*> weight. In the pooled dataset, all subjects were given with a
*

*> single dose of that drug. Part of the control stream was
*

*> provided as follows. Based on the prior knowledge that a
*

*> two-compartment model with first-order absorption and
*

*> elimination is used to describe its PK profile well, a
*

*> two-compartment model was used for my meta-popPK analysis.
*

*> After the NM run for the additive or proportional error model
*

*> or both, the NM indicated that "*MIN*IMIZATION *T*ERMINATED
*

*> DUT TO ROUNDING ERRORS (ERROR=134)" alone or in combination
*

*> with "ADDITIONAL PROBLEMS OCCURED WITH THE MINIMIZATION.
*

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

*> ACCEPT THEM ONLY AFTER CHECKING THE COVARIANCE STEP PRODUCES
*

*> REASONABLE OUTPUT".
*

*>
*

*> In addition, the typical values of the relative
*

*> bioavailability (*F1*) was estimated as 5 - 8 for STDY 3 (that
*

*> is, f13), 9 -14 for STDY 5 (or f15) and 0.4 - 0.53 for STDY 6
*

*> (f16). Obviously, only f16 seems to be reasonable. The drug
*

*> used in the STDY 3 and 5 was given extravascularly, and thus
*

*> their F1 values should be less than 1. To fix these problems,
*

*> I have made many efforts but failed. For this, I would
*

*> appreciate your time and suggestions.
*

*> Thank you,
*

*> Hong-Guang Xie
*

*> ==================
*

*> ............
*

*> $SUBROUTINE ADVAN4 TRANS4
*

*>
*

*> $PK
*

*> ETAMX = 10
*

*>
*

*> SI1=0 SI2=0
*

*> SI3=0
*

*> SI4=0
*

*> SI5=0
*

*> SI6=0
*

*>
*

*> IF (STDY .EQ. 1) SI1=1 ; iv
*

*> IF (STDY .EQ. 2) SI2=1 ; iv
*

*> IF (STDY .EQ. 3) SI3=1 ; epidural
*

*> IF (STDY .EQ. 4) SI4=1 ; iv
*

*> IF (STDY .EQ. 5) SI5=1 ; rectal
*

*> IF (STDY .EQ. 6) SI6=1 ; oral
*

*>
*

*> IV = SI1+SI2+SI4 ;intravascular dosing
*

*>
*

*> ;APPARENT CLEARANCE
*

*> TVCL=popcl
*

*> PPVCL=bsvcl
*

*> IF (ABS(PPVCL) .GE. ETAMX) EXIT 1 10 CL=TVCL*EXP(PPVCL)
*

*> ;CENTRAL VOLUME OF DISTRIBUTION
*

*> TVV2=popv2
*

*> PPVV2=bsvv2
*

*> IF (ABS(PPVV2) .GE. ETAMX) EXIT 1 20
*

*> V2=TVV2*EXP(PPVV2)
*

*>
*

*> ;INTERCOMPARTMENTAL CLEARANCE
*

*> TVQ=popq
*

*> PPVQ=bsvq
*

*> IF (ABS(PPVQ) .GE. ETAMX) EXIT 1 30
*

*> Q=TVQ*EXP(PPVQ)
*

*>
*

*> ;PERIPHERAL VOLUME OF DISTRIBUTION
*

*> TVV3=popv3
*

*> PPVV3=bsvv3
*

*> IF (ABS(PPVV3) .GE. ETAMX) EXIT 1 40
*

*> V3=TVV3*EXP(PPVV3)
*

*>
*

*> ;FIRST ORDER ABSORPTION RATE CONSTANT
*

*> TVKA=SI3*ka3+SI5*ka5+SI6*ka6
*

*> PPVKA=bsvka IF (ABS(PPVKA) .GE. ETAMX) EXIT 1 50
*

*> KA=TVKA*EXP(PPVKA)
*

*>
*

*> ;LAG TIME
*

*> TVLAG=SI3*lag3+SI5*lag5+SI6*lag6
*

*> PPVLAG=bsvlag IF (ABS(PPVLAG) .GE. ETAMX) EXIT 1 60
*

*> ALAG1=TVLAG*EXP(PPVLAG)
*

*>
*

*> ;BIOAVAILABILITY (BA)
*

*> TVF1=SI3*f13+SI5*f15+SI6*f16
*

*> PPVF1=bsvf1
*

*> IF (ABS(PPVF1) .GE. ETAMX) EXIT 1 70
*

*> BA=TVF1*EXP(PPVF1)
*

*> IF (IV .GE. 1) THEN
*

*> F1=1
*

*> ELSE
*

*> F1=BA
*

*> ENDIF
*

*>
*

*> ;ZERO ORDER ABSORPTION DURATION
*

*> TVD2=SI1*d21+SI2*d22+SI4*d24
*

*> PPVD2=bsvd2
*

*> IF (ABS(PPVD2) .GE. ETAMX) EXIT 1 80
*

*> D2=TVD2*EXP(PPVD2)
*

*>
*

*> RMIN=AMT/(60*D2) ;mcg/minute
*

*> ; SCALING FOR CENTRAL COMPARTMENT (I.E., CONCENTRATION
*

*> COMPARTMENT)
*

*> S2=V2 ; DOSE IN MCG, CONCENTRATION IN NG/ML=MCG/L
*

*>
*

*> $ERROR
*

*> IPRED = F
*

*>
*

*> IF (MDV .EQ. 0) THEN
*

*> W = SQRT(add**2 + F*F*prop**2)
*

*> ELSE
*

*> W = 1
*

*> ENDIF
*

*>
*

*> IRES=DV-IPRED
*

*>
*

*> IF (W .EQ. 0) THEN
*

*> IWRES=IRES
*

*> ELSE
*

*> IWRES=IRES/W
*

*> ENDIF
*

*>
*

*> Y=F+W*eps1*EXP(bsvres)
*

*>
*

*> $THETA (0,11.5) ;popcl
*

*> (0,24.1) ;popv2
*

*> (0,12.9) ;popq
*

*> (0,8.22) ;popv3
*

*> (0,0.14) ;ka3
*

*> (0,0.109) ;ka5
*

*> (0,0.397) ;ka6
*

*> (0 FIX) ;lag3
*

*> (0 FIX) ;lag5
*

*> (0 FIX) ;lag6
*

*> (0,4.93) ;f13
*

*> (0,9.19) ;f15
*

*> (0,0.523) ;f16
*

*> (0,0.00072) ;d21
*

*> (0,0.00007) ;d22
*

*> (0,0.00003) ;d24
*

*> (0, 0.01) ;add
*

*> (0, 0.05) ;prop
*

*> $OMEGA 0.828 ;bsvcl
*

*> $OMEGA 1.63 ;bsvv2
*

*> $OMEGA 0 FIX ;bsvq
*

*> $OMEGA 0 FIX ;bsvv3
*

*> $OMEGA 0 FIX ;bsvka
*

*> $OMEGA 0 FIX ;bsvlag
*

*> $OMEGA 0 FIX ;bsvf1
*

*> $OMEGA 0 FIX ;bsvd2
*

*> $OMEGA 0 FIX ;bsvres
*

*> $SIGMA 1. FIX ; eps1
*

*>
*

*> $ESTIMATION NOABORT MAXEVAL=9999 POSTHOC PRINT=5 SIGDIGITS=3 MSFO=
*

*> METHOD=CONDITIONAL INTERACTION
*

*>
*

*> ;$COV PRINT=E
*

*>
*

*> $TABLE ID STDY TIME DV IPRED AMT MDV CMT CL V2 Q V3 KA ALAG1
*

*> F1 D2 RATE RMIN ETA1 ETA2 ETA3 ETA4 ETA5 ETA6 ETA7 ETA8 ETA9
*

*> IWRES IRES WT BWT AGEY AGED GAGE
*

*> NOPRINT ONEHEADER FILE=
*

*> ===================================================================================
*

*>
*

*>
*

*> --
*

*> Nick Holford, Dept Pharmacology & Clinical Pharmacology
*

*> University of Auckland, 85 Park Rd, Private Bag 92019, Auckland,
*

*> New Zealand
*

*> n.holford *

*> tel:+64(9)373-7599x86730 fax:+64(9)373-7090
*

*> http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
*

*>
*

*>
*

*>
*

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand

n.holford

http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Received on Thu Aug 21 2008 - 19:00:55 EDT

Date: Fri, 22 Aug 2008 11:00:55 +1200

Hong-Guang,

It seems you are already aware of the publications that have

demonstrated that NONMEM's warnings and failure to run the $COV step are

not of much use for model evaluation. So please stop worrying about them

and focus on more important things e.g. is your data correct (as Mark

Sale suggested).

You wrote:

"If this is the case at any time or even at most time, the NONMEM run

would not seem to be too tough."

I dont really understand what you mean. If you mean that using NONMEM is

much easier if we ignore these termination warnings then I agree with

you. Pay attention to your data and the model predictions. Dont pay

attention to these warnings that do not have any useful relationship to

the purpose of modelling.

Nick

Hong-Guang Xie wrote:

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand

n.holford

http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Received on Thu Aug 21 2008 - 19:00:55 EDT