M. de Hoog1,2,, A.A.T.M.M. Vinks4, R.C. Schoemaker2,3, J.N. van den Anker1,2
Department of Pediatrics1, Erasmus University and University Hospital Rotterdam/Sophia Children Hospital, Rotterdam, The Netherlands, Pediatric Pharmacology Network2, Leiden, The Netherlands, Center for Human Drug Research3, Leiden, The Netherlands and TDM & Clinical Toxicology Laboratory/The Hague Central Hospital Pharmacy4, The Hague, The Netherlands.
Introduction: Nonlinear mixed effect modeling (NONMEM) and nonparametric expectation maximization (NPEM) have both been used in population pharmacokinetic modeling of intravenous antibiotics. NONMEM assumes a normal distribution of pharmacokinetic parameters over patients, whereas NPEM makes no assumption about the distribution other than a limit to the possible values. Both models give as result population pharmacokinetic parameters with a standard deviation and indiviudal posterior estimates. A preliminary comparison of these methods has been described once. We compared both methods with respect to differences in population pharmacokinetic parameters and performance in explaining the original patient data set.
Methods: A retrospective study was performed in 470 neonates who had received tobramycin with a gestational age dependent interval and dose of 12-24 hours and 2.5-3.5 mg/kg, with trough and peak serum levels taken before and after the fourth dose . Data were analyzed according to a one-compartment open model with NONMEM V. Residual variability was described using a constant coefficient of variation error model. The same patient group was described using NPEM, with a residual variability model of SD = 0.0599 + 0.0126.C+0.00438.C2 based on the serum assay error pattern. Individual pharmacokinetic parameter estimates of both models were compared.
Results: Coefficient of variation for the residual error for the NONMEM model was 21%. If the residual variability for the NPEM analysis is approximated by a straight line, a residual variability in the order of 5 to 6 % would result.
Mean posterior parameter values for the respective models were: Ke-NONMEM:0.072 hr-1 (range: 0.031-0.127, SD:0.016), Ke-NPEM: 0.079 hr-1 (range: 0.001-0.24, SD: 0.033), Vd-NONMEM: 0.60 l/kg (range: 0,542-0,636, SD: 0.02), Vd-NPEM: 0.64 l/kg (range: 0.28-2.25, SD: 0.30). Range of Ke and especially Vd for NPEM was much larger than for NONMEM. Ke-NPEM and Ke-NONMEM were strongly correlated (p<0.001) while Vd-NONMEM and Vd-NPEM were not correlated. The larger range in parameter values for the NPEM analysis results in a closer description of the measurements than with NONMEM parameters. This is related to the fact that NPEM attaches more credibility to the individual data points than NONMEM; the residual error for NPEM is assumed to be much smaller than for NONMEM.
Conclusions:
1. NONMEM and NPEM give similar results in mean population parameter estimates.
2. Individual NPEM Vd and Ke estimates show a much larger range than NONMEM, due to the fact that NPEM model predictions follow the data more closely than NONMEM predictions.
Reference: PAGE 9 () Abstr 125 [www.page-meeting.org/?abstract=125]
Poster: poster