2.5 Population pharmacokinetic analysis
Analysis of PPK were performed by nonlinear mixed-effects modeling
software (NONMEM version 7.4.3 ICON Development Solutions, Ellicott
City, MD, USA). The first-order conditional estimation with interaction
(FOCE-I) option was chosen to evaluate pharmacokinetic parameters and
variability. One- and two-compartment structural kinetic models with
first-order and Michaelis–Menten elimination were compared during the
base model determination step. An exponential model was chosen to
describe the inter-individual variability of voriconazole
pharmacokinetic parameters. Furthermore, additive, exponential,
proportional and combined residual models were used to evaluate the
residual variability.
We initially examined the correlation between pharmacokinetic parameters
and potential covariates (associated with voriconazole
Cmin in univariate analysis) by visualizing linear plots
and box plots. Subsequently, a stepwise forward and backward exclusion
approach was used to develop the covariate model. Covariates were
considered significant in the case where the following conditions
occurred simultaneously: inclusion of covariates resulted in a decrease
in the objective function value (OFV) greater than 3.84 (p<
0.05) and exclusion of covariates resulted in an increase in OFV greater
than 6.64 (p< 0.01).
Model evaluation was carried out by Goodness of fit plots (GOFs), visual
predictive check (VPC), normalized prediction distribution error (NPDE)
and nonparametric bootstrapping analysis (Bootstrap). The GOFs were used
to evaluate the adequacy of fitting. The VPC test was performed by 1000
simulations based on the final model to assess the predictive
performance of the final model. NPDE could estimate the predictive
characteristics of the final model by statistical tests and diagnostic
charts. Bootstrap is used to evaluate the robustness and stability of
the final model by repeating the original data 1000 times with different
random samples.