John C. Lukas1, George L. Drusano2, Paolo Vicini1
1) Resource Facility for Population Kinetics, Department of Bioengineering, University of Washington, Seattle, WA, 2) Division of Clinical Pharmacology, Department of Medicine and Pharmacology, Albany Medical College, NY
The population pharmacokinetics (PK) of the l-enantiomer of ofloxacin (levofloxacin) (data published in Preston e t al., Antimicrob. Agents Chemother., 42, 1998) were studied using a two-compartment PK model with NONMEM (First Order method). Up to three infusion or oral doses were administered to 272 patients with varying infection sites and sampled sparsely (2-6 times) after the last dose. Typical population PK parameter estimates were for clearance (mean (interindividual CV%)), Cl=7.63 L/h (57), volume of distribution, V=53.50 L (59), central to peripheral compartment transfer rate, k12=0.34 h-1 (110) and peripheral to central compartment rat e, k21=0.39 h-1 (52). The transfer rates both had lognormal distributions. Age was (mean ± SD) 46 ± 18 years and weight 78 ± 18 Kg. The Bayesian posthoc PK parameter estimates from an index group of 172 patients were regressed against 6 covariates (age, weight, ra ce, sex, creatinine clearance and site of infection) using hierarchical forward inclusion of covariates in distinct NONMEM runs. The covaria te model established for Cl was based on age and creatinine clearance, and that for V on weight and age. These regressions were then used to predict Cl and V in the rest of the population. Bias (mean (95% C.I.)) for covariate model predictions vs. Bayesian posthoc estimates were 4.99% (-1.73%, 16.71%) for Cl and 3.18% (-2.52%, 8.88%) for V, and precision was 32.4% and 27.5% respectively. The above results have been compared to those of the same procedure obtained with a nonparametric population analysis package (NPEM2, LAPK, USC) employing maximum a pos teriori probability Bayesian estimates of the PK parameters and a backward regression of those on the covariates. Overall the NONMEM and NPE M2 approaches perform similarly. There was no significant difference in the empirical (n=272, no covariates) PK parameter estimates from the two methods (p>0.14, unpaired t-test). Interestingly however, the covariate models determined with NONMEM were reduced versions, nested within those obtained in the NPEM2 approach. The development of covariate models for parameters, particularly for sparse datasets such as t his one, via a forward inclusion in successive NONMEM runs proves to be a powerful and robust method. The optimum model showed a 23% reducti on in the first order objective function (p<0.00001) compared to the base model (no covariates).
Reference: PAGE 9 (2000) Abstr 97 [www.page-meeting.org/?abstract=97]
Poster: poster