E I Ette1, T M Ludden. 1. Division of Biopharmaceutics, Center for Drug Evaluation and Research, Food and Drug Administration, Rockville, Maryland 20857, USA.
Abstract
PURPOSE: The usefulness of several modelling methods were examined in the development of a population pharmacokinetics model for cefepime. METHODS: The analysis was done in six steps: (1) exploratory data analysis to examine distributions and correlations among covariates, (2) determination of a basic pharmacokinetic model using the NON-MEM program and obtaining Bayesian individual parameter estimates, (3) examination of the distribution of parameter estimates, (4) multiple linear regression (MLR) with case deletion diagnostics, generalized additive modelling (GAM), and tree-based modelling (TBM) for the selection of covariates and revealing structure in the data, (5) final NONMEM modelling to determine the population PK model, and (6) the evaluation of final parameter estimates. RESULTS: An examination of the distribution of individual clearance (CL) estimates suggested bimodality. Thus, the mixture model feature in NONMEM was used for the separation of subpopulations. MLR and GAM selected creatinine clearance (CRCL) and age, while TBM selected both of these covariates and weight as predictors of CL. The final NONMEM model for CL included only a linear relationship with CRCL. However, two subpopulations were identified that differed in slope and intercept. CONCLUSIONS: The findings suggest that using informative graphical and statistical techniques enhance the understanding of the data structure and lead to an efficient analysis of the data.
PURPOSE: The usefulness of several modelling methods were examined in the development of a population pharmacokinetics model for cefepime. METHODS: The analysis was done in six steps: (1) exploratory data analysis to examine distributions and correlations among covariates, (2) determination of a basic pharmacokinetic model using the NON-MEM program and obtaining Bayesian individual parameter estimates, (3) examination of the distribution of parameter estimates, (4) multiple linear regression (MLR) with case deletion diagnostics, generalized additive modelling (GAM), and tree-based modelling (TBM) for the selection of covariates and revealing structure in the data, (5) final NONMEM modelling to determine the population PK model, and (6) the evaluation of final parameter estimates. RESULTS: An examination of the distribution of individual clearance (CL) estimates suggested bimodality. Thus, the mixture model feature in NONMEM was used for the separation of subpopulations. MLR and GAM selected creatinine clearance (CRCL) and age, while TBM selected both of these covariates and weight as predictors of CL. The final NONMEM model for CL included only a linear relationship with CRCL. However, two subpopulations were identified that differed in slope and intercept. CONCLUSIONS: The findings suggest that using informative graphical and statistical techniques enhance the understanding of the data structure and lead to an efficient analysis of the data.
Authors: P Burtin; E Jacqz-Aigrain; P Girard; R Lenclen; J F Magny; P Betremieux; C Tehiry; L Desplanques; P Mussat Journal: Clin Pharmacol Ther Date: 1994-12 Impact factor: 6.875