Re: Modeling compartment initial conditinns using ETA()= EPS(1) and CPZERO measure
I don't understand Kazimierz's model, but I can answer his question:
$ERROR
COM(1)=ERR(1)
;QUESTION: WHY ERR(1)=0.0 in write-file and in TABLE?
Epsilons are always zero except during the simulation step.
E.g., with Conditional Estimation, ETA is non-zero, but EPS is zero.
Only the partial derivatives of Y wrt. EPS enter into the objective
function.
He needs to use EPS in the Error model.
E.g., assume that every subject's data starts with an observation record
at TIME 0 giving
the observed value of the initial steady state concentration, and that
ISSC is the predicted value.
(ISSC might be predicted as in Nick's e-mail, or via SS dose, as in the
"endogenous supplementation example" help item, or other ways - I don't
want to get into a discussion of this issue right now.)
ISSC depends on theta and eta and covariates, not ERR(1).
Then a possible model is:
$ERROR
ISSC= ... whatever ...
Y1=ISSC+ERR(1) ; or other error model for initial SS observations
Y2=F+ERR(2) ; or other error model for remainder of observations
IF (TIME.EQ.0) THEN
Y=Y1
ELSE
Y=Y2
ENDIF
On Thu, 11 Oct 2007 19:07:14 +0200, "Kazimierz H. Kozlowski"
<[EMAIL PROTECTED]> said:
> Dear NM-Users,
>
> I need a method to force using EPS(1) intead ETA() for estimating
> initial steady-state compartment concentration for $DES.
> Pre-dose TIME for SS-ending is known. I used the following
> abbreviated codes in ERROR, and NONMEM act well, but predics individual
> regression stricted to measure CPZERO.
>
> sincerely
> Kazimierz H. Kozlowski
>
> $DES
> DADT(1)=-K*A(1)-K12*A(1)+K21*A(2)
> DADT(2)=K12*A(1)-K21*A(2)
> $ERROR
> COM(1)=ERR(1) ;QUESTION: WHY ERR(1)=0.0 in write-file and in
> TABLE?
> WRITE (50,*) COM(1),ERR(1),ICALL,COMACT ; COM(1)=0, ICALL=2 always
> FZ=THETA(10)*ERR(1)*THETA(10)*ERR(1) ;FZ=ERR**2
> FZ1=1.0-FZ*THETA(9)*THETA(9) ;FZ1=1-ERR*2*TH9*2
> EXP1=(K-BETA)*EXP(-ALPHA*(IAGE+TIME))
> EXP2=(ALPHA-K)*EXP(-BETA*(IAGE+TIME))
> EXP3=ALPHA*EXP(-BETA*(IAGE+TIME))
> EXP4=BETA*EXP(-ALPHA*(IAGE+TIME))
> EN=(C01-(C01*C01-FZ1*(C01*C01-FZ))**0.5)/FZ1;EN=ENDOG. CP(0)
> CS=EN*(ALPHA-BETA)
> ;CSS=EN(T=-IAGE)
> CSS=CS/((K-BETA)*EXP(-ALPHA*IAGE)+(ALPHA-K)*EXP(-BETA*IAGE))
> IPR1=F+CSS/(ALPHA-BETA)*(EXP1+EXP2) ;PLASMA CONC.
> IPR2=A(2)/V1*K21/K12+CSS/(ALPHA-BETA)*(EXP3-EXP4);TISSUE CONC.
>
> RZ=CSS/CLE
> ;ENDOG. RATE
> IDIF=CP-IPR1
> W=(1+IPR1*IPR1*THETA(9)*THETA(9))**0.5
> Y=IPR1+W*THETA(10)*ERR(1)
> $THETA
> .
> .
> .
> $THETA
> $SIGMA 1
> FIXED
> ;VARIANCE FOR ERR(1)
> $EST NOABORT NUMERICAL SLOW METHOD=1 INTERACTION LAPLACIAN POSTHOC
> SIG=6 MAX=9999 PRINT=1 MSFO=VER1.MSF
> $COVARIANCE MATRIX=S SLOW COMPRESS PRINT=E
> $TABLE ID TIME ALPHA CLD CLE BETA V1 RZ NOPRINT FILE=VER1.TAB
> $TABLE ID TIME C01 ER=COM(1) EN CSS IPR1 IPR2 NOPRINT FILE=VER1.TAB
> $TABLE ID TIME IAGE K12 K21 K R1 IDIF NOPRINT FILE=VER1.TAB
> $SCAT CP VS IPR1 UNIT
>
--
Alison Boeckmann
[EMAIL PROTECTED]