I have started using NONMEM-V for population pharmacokinetics. I have been facing some problem in specifying a negative slope to any regression equation (like relationship between clearance and age). The output message says a nonpositive value for clearance and aborts the run. I tried both ways: (1) introducing (-) sign in the equation itself and (2) giving the negative value for lower constraint for the slope (THETA). The output message remains the same. Is it due to the large variability in the data or there is some other way of solving this problem. I should mention that the installation was done through SETUP.BAT (shortcut method). It would be nice if you could suggest me to solve this problem.
Thanks for the help.
Post Doc. Res. Assoc.
Dept. of Pharmaceutics,
College of Pharmacy,
University of Florida,
Gainesville, FL 32610
From: "Jogarao V. Gobburu" <GOBBURUJ@gunet.georgetown.edu>
Subject: NONMEM -Reply
Date: 9 Jul 1998 12:44:42 -0400
One problem could be the way you set your ETA for CL. If the ETA for CL is assumed to be additive (normal distr.), then it is finitely possible to encounter negative values of CL. A way of avoiding it is by using a proportional ETA: CL=THETA()*EXP(ETA()).This may be more meaningful too, in that, PK/PD parameters generally are log-normally distributed. I hope this helps.
Center for Drug Development Science,
Georgetown Univ., Washington DC.
From: email@example.com.EDU (ABoeckmann)
Subject: Nonpositive clearance
Date: 9 Jul 1998 16:05:55 -0400
Nagaraja does not give any details, but may have a model like this:
TVCL = THETA(1) - THETA(2)*AGE ; lower bound of theta(2) = 0
There is no straightforward way to specify an upper bound for theta(2)that prevents TVCL from becoming negative for large AGE and small THETA(1). With the model
CL = THETA(1) - THETA(2)*AGE + ETA(1)
the difficulty is even worse, because now a negative eta with large absolute value will also cause trouble.
Lewis Sheiner suggests:
LOGCL = THETA(1) - THETA(2)*AGE + ETA(1)
CL = EXP(LOGCL)
Now CL can never be negative. Note that this is equivlent to
LOGTVC = THETA(1) - THETA(2)*AGE
CL = EXP(LOGTVC) * EXP(ETA(1))
so that the eta is really exponential, not additive.