Dear Zahid:

It is not enough to say you are using WinNonlin. What specific method are you using to fit your data? What weighting scheme are you using to describe the credibility of your data? I would strongly suggest that you weight your data by the reciprocal of the assay variance at each measurement. There are easy ways to do this, but the laboratory community still thinks that CV% is the correct measure of assay precision. They are totally wrong. You might look at

Jelliffe RW, Schumitzky A, Van Guilder M, Liu M, Hu L, Maire P, Gomis P, Barbaut X, and Tahani B: Individualizing Drug Dosage Regimens: Roles of Population Pharmacokinetic and Dynamic Models, Bayesian Fitting, and Adaptive Control. Therapeutic Drug Monitoring, 15: 380-393, 1993.

In addition, most 2 compartment models usually only have observations made from the central, serum, compartment. You cannot compute the volume of an unobserved compartment without making further assumptions, which is what is usually done in the situations you describe. On the other hand, one can always calculate the amounts of drug in the central and peripheral compartments by fitting serum concentrations alone, but the volume can never be computed without making further assumptions such as that of assuming that, at the steady state, the rates of transfer are equal between the 2 compartments, and therefore that the clearance in both directions must be the same. From this one then calculated the apparent volume in the unobserved peripheral compartment. That is why hey always do this only for an assumed steady state.

Also, you might well consider using the approach taken by Dr. Michael Neely in his Pmetrics software, which has unconstrained parameter distributions. In contrast to those approaches which simply assume that the model parameter distributions have normal or lognormal shapes, the Pmetrics software makes no such assumptions. Because of this, results with Pmetrics are always more likely given the data, than those which do, as parametric approaches assume and constrain the model parameter distributions to be normal, lognormal, or some other assumed distribution. Pmetrics does not do this, and it has been shown that the results given the data are always more likely as a result.

See Bustad A, Terziivanov D, Leary R, Port R, Schumitzky A, and Jelliffe R: Parametric and Nonparametric Population Methods: Their Comparative Performance in Analysing a Clinical Data Set and Two Monte Carlo Simulation Studies. Clin. Pharmacokinet., 45: 365-383, 2006.

You might also look at Neely M, van Guilder M, Yamada W, Schumitzky A, and Jelliffe R: Accurate Detection of Outliers and Subpopulations with Pmetrics, a Nonparametric and Parametric Pharmacometric Modeling and Simulation Package for R. Therap. Drug Monit. 34: 467-476, 2012.

I hope these references and comments might be helpful to you. All the best for the Holidays!

Roger Jelliffe