From: Nick Holford <*n.holford*>

Date: Thu, 03 Jul 2008 20:03:11 +0200

Huali,

NONMEM's failure to compute standard errors says nothing about the

quality of your model. This has been discussed at length on nmusers. You

should look at the parameter estimates for feasibility and use

simulation based diagnostics e.g. VPC, to evaluate your model fit. No

need to waste time worrying about lack of SEs. Holford's rule -- the

covariance step runs with simple models. When the model gets interesting

NONMEM will stop calculating SEs.

Nick

Huali Wu wrote:

*> Dear NMusers:
*

*>
*

*> I am trying to fit a data set with concentrations of encapsulated drug
*

*> and released drug. These concentrations were measured after an hour
*

*> infusion of encapsulated drug. The problem is that I cannot get
*

*> standard error estimated from simultaneous fitting of all the data.
*

*> And I got an error message as below:
*

*>
*

*> 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.
*

*> NO. OF FUNCTION EVALUATIONS USED: 968
*

*> NO. OF SIG. DIGITS IN FINAL EST.: 3.2
*

*> R MATRIX ALGORITHMICALLY NON-POSITIVE-SEMIDEFINITE
*

*> BUT NONSINGULAR
*

*> R MATRIX IS OUTPUT
*

*> COVARIANCE STEP ABORTED
*

*>
*

*> But if I fit the encapsulated drug first and then fit data of both
*

*> encapsulated drug and released drug by fixing the typical values of
*

*> parameters associated with encapsulated drug compartment, I can get an
*

*> estimation of standard errors. And GOF plots look good.
*

*>
*

*> So my question is can I use this stepwise strategy to build up my base
*

*> model? If not, what can I do to get standard error estimated in
*

*> simultaneous fitting?
*

*>
*

*> Any comment or suggestion will be highly appreciated.
*

*>
*

*> Below is my code:
*

*>
*

*> $SUBROUTINES ADVAN9 TRANS1 TOL=3
*

*> $MODEL NPAR=8, NCOMP=3, COMP=(CENTRAL,DEFOBS),
*

*> COMP=(PERIPH1), COMP=(PERIPH2)
*

*>
*

*> $PK
*

*> V1 = THETA(1)*EXP(ETA(1))
*

*> V2 = THETA(2)*EXP(ETA(2))
*

*> V3 = THETA(3)
*

*> VM = THETA(4)
*

*> KM = THETA(5)
*

*> Q2 = THETA(6)*EXP(ETA(3))
*

*> Q3 = THETA(7)
*

*> Q4 = THETA(8)
*

*> S1 = V1
*

*> S2 = V1
*

*>
*

*> $ERROR
*

*>
*

*> R1=0
*

*> IF (CMT.EQ.1) R1=1
*

*> R2=0
*

*> IF (CMT.EQ.2) R2=1
*

*> Y1=F+F*EPS(1)
*

*> Y2=F+F*EPS(1)
*

*> Y=R1*Y1+R2*Y2
*

*> IPRED=F
*

*> IRES=DV-IPRED
*

*>
*

*> $DES
*

*> C1 = A(1)/V1
*

*> DADT(1) = - C1*VM/(KM+C1)-Q2*A(1)/V1
*

*> DADT(2) = C1*VM/(KM+C1)+Q2*A(1)/V1 + Q4*A(3)/V3-
*

*> Q3*A(2)/V2-Q4*A(2)/V2
*

*> DADT(3) = Q4*A(2)/V2-Q4*A(3)/V3
*

*>
*

*> $THETA (5.27 FIX) (0, 0.5) (0, 3.38) (0.329 FIX) (4.38 FIX) (0.136
*

*> FIX) (0, 0.0626) (0, 0.0461)
*

*> $OMEGA (0.01) (0.01) (0.01)
*

*> $SIGMA (.1)
*

*>
*

*> $EST MAXEVAL=9999 PRINT=5
*

*> $COV
*

*> $TABLE ID TIME DV AMT RATE CMT NOPRINT FILE=ENCAP4 ONEHEADER
*

*> $SCAT (RES WRES) VS TIME BY ID
*

*>
*

*> Best regards,
*

*>
*

*> Huali
*

*>
*

*>
*

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

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

n.holford

www.health.auckland.ac.nz/pharmacology/staff/nholford

Received on Thu Jul 03 2008 - 14:03:11 EDT

Date: Thu, 03 Jul 2008 20:03:11 +0200

Huali,

NONMEM's failure to compute standard errors says nothing about the

quality of your model. This has been discussed at length on nmusers. You

should look at the parameter estimates for feasibility and use

simulation based diagnostics e.g. VPC, to evaluate your model fit. No

need to waste time worrying about lack of SEs. Holford's rule -- the

covariance step runs with simple models. When the model gets interesting

NONMEM will stop calculating SEs.

Nick

Huali Wu wrote:

--

Nick Holford, Dept Pharmacology & Clinical Pharmacology

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

n.holford

www.health.auckland.ac.nz/pharmacology/staff/nholford

Received on Thu Jul 03 2008 - 14:03:11 EDT