Objectives: The objectives of this study were to evaluate the pharmacokinetics of voriconazole in liver transplant patients, probe covariate effects on voriconazole pharmacokinetics, externally validate the model and explore limited sampling strategies (LSSs) using Bayesian approaches.
Methods: Full pharmacokinetic profiles were collected within one oral dosing interval from 13 liver transplant patients. Nonlinear mixed-effects pharmacokinetic models were developed using NONMEM software. The final model was internally evaluated using bootstrapping and visual predictive check (VPC), and externally validated by predicting additional samples from different patients that were not used for model building. Maximum a posteriori Bayesian estimators were developed to predict the area under the plasma concentration-time curve (AUC) using the validated final model as the a priori model, actual dosing record and covariate values as the input, and a few concentrations (limited sampling) as feedback information (LSS). Mean prediction error (MPE) and mean absolute prediction error (MAPE) were calculated for external validation and LSS.
Results: A one-compartment model with an absorption lag time (t(lag)) adequately described the data. Population estimates of total clearance after oral administration (CL/F) and volume of distribution after oral administration (V(d)/F) were 7.92 L/h and 248 L, respectively. Values of CL/F, V(d)/F and t(lag) decreased with post-operative time and converged to stable levels in about 7 post-operative days. CL/F significantly decreased with increased international normalized ratio. Co-administration of pantoprazole, race and alanine aminotransferase were also significantly associated with pharmacokinetic parameters but ultimately excluded in the final model. VPC showed that most of the data fell within the 90% prediction interval and were symmetrically distributed around the median. Additional 52 samples from 19 patients were collected for external validation. MPE was 0.206 μg/mL (not significantly different from zero) and MAPE was 0.99 μg/mL. Compared with trough levels, LSS using two samples or one sample at a different time provided better MPE, MAPE and correlation (R2) between the observed and LSS-predicted AUC.
Conclusions: The population model that was developed showed significant association of voriconazole pharmacokinetics with post-operative time and liver function, and was able to predict an independent external dataset. Our observations suggested a need for intravenous administration of voriconazole in the immediate post-operative period before an oral dose can be administrated. LSS using one sample appeared to be sufficient for reasonable AUC estimation.