From: "atul"
Subject:[NMusers] Oral pharmacokinetics sparse (no absorption phase)
Date: Tue, 7 May 2002 23:23:21 -0700
Hello nmusers
I am sure this issue might have been discussed before but I was not able to find it.
Could you share your views on it?
I am analysing a real sparse data (only 4 points at 2, 4, 8 and 12h) after oral administration
at steady state. So no points in the absorption phase at all. Upon examination of the data
I saw that a one compartment model is good enough. So I went ahead and did analysis in ADVAN2.
I have previous values from a paper reporting a ka of 2.8 for my drug. If I fix initial estimate
of Ka as 2.8 and try to give eta on ka then NONMEM gives me error message 1. If I give the
initial estimate of ka as 2.8 (without eta) and do not fix it then still I get the error message 1.
Now when I change the initial estimate of ka to 1.3 surprisingly the run converges (everything
is same as before with inital estimate as 2.8). I get final estimates of CL, V and ka as 233, 1240
and 1.13. I then changed the initial estimate of ka to 0.3. I then get the final estimates of CL, V
and ka as 251, 211 and 0.174. The estimate of V (211) is similar to what has been reported earlier.
However, The PRED vs DV plots are much better with higher ka. I get totally different meaningless
values if I choose ka as 0.8. I think the ratio (ka/V) is really which is affecting the run.
From what I observe is that the whole analysis is critically dependent on the initial value of ka which
I choose. In case of ka=0.3 I end up with a flip-flop situation.
How can I be sure of the final estimates which I get from these results? How would one decide in
these situations? Is there any reference which
discusses this.
LIST OF ERROR MESSAGES:
1. ERROR IN NCONTR WITH INDIVIDUAL 13 ID= .19000000E+02
NUMERICAL HESSIAN OF OBJ. FUNC. FOR COMPUTING CONDITIONAL ESTIMATE
IS NON POSITIVE DEFINITE
MESSAGE ISSUED FROM ESTIMATION STEP
AT 0TH ITERATION, UPON EVALUATION OF GRADIENT WITH RESPECT TO OMEGA
0TRY SUBSTANTIALLY INCREASING "INITIAL ESTIMATE OF SIGMA"
2. Elimination from some compartment is negligible. Steady state not achievable.
3. MATH overflow error.
Thanks in advance for your time
Venaktesh Atul Bhattaram
Post-doctoral Fellow
University of Florida
Gainesville-32610
Oral pharmacokinetics sparse (no absorption phase)
From: Leonid Gibiansky
Subject: Re: [NMusers] Oral pharmacokinetics sparse (no absorption phase)
Date: Wed, 08 May 2002 09:30:34 -0400
Atul,
If you compare two solutions given to you by NONMEM (S+ script for population curves
is given below) you will see that they cannot be distinguished even given the dense PK
data (they are nearly identical except 10% difference in Cmax). You need IV data or
previous knowledge to sort this out.
Leonid.
From: "Bachman, William"
Subject: RE: [NMusers] Oral pharmacokinetics sparse (no absorption phase)
Date: Wed, 8 May 2002 11:55:11 -0400
Atul,
One might try a formal sensitivity test using a range of fixed ka's (you've done this in
a limited informal way already).
1. pick the range and interval of fixed ka values to test and make the runs
2. compare the effect of changing ka on the remaining parameter estimates and their
standard errors, goodness of fit (plots, obj. fcn values, and variance estimates)
(In all cases assure yourself you have a global minimum, successful $COV step, etc.)
3. pick the "best" overall fixed ka and live with it.
Bill
*******