Population pharmacokinetic modelling
GBP concentration-time data were analysed using the nonlinear mixed-effects modelling software MONOLIX (Version 2019, Lixoft, France). Model development was based in 4 steps: selection of structural model; selection of an error model; covariate analysis and final model internal validation. The model selection was guided by the objective function value (OFV), represented as -2 times the log-likelihood (-2LL), relative standard errors (RSE) below 40% and unbiased goodness-of-fit plots [49-51]. One and two-compartment models, the inclusion of lag time and linear or Michaelis-Menten elimination were tested. A log-normally distributed interindividual variability (IIV) was tested on model parameters, whereas proportional, additive and combined error models were tested.
The following continuous covariates were evaluated: body weight, height, BMI, eGFR, HbA1c and FGL. Categorical covariates included sex, OCT2 and OCTN1 genotypes and the diabetes mellitus diagnostics. Potential covariates were evaluated by Pearson’s correlation test or ANOVA. Covariates were included in the model using forward inclusion and backward elimination with a level of significance of p < 0.05 (ΔOFV > -3.84 points) and p < 0.01 (ΔOFV < 6.63), respectively. Additionally, the covariate inclusion had to reduce the unexplained IIV and improve the goodness-of-fit plots [49-51].
The model was internally validated using visual predictive check (VPC, n = 10.000 simulations) and bootstrap (n = 5.000 replicates) analysis, using the package Rsmlx for RStudio software (version 1.1.442, Free Software Foundation, Boston, USA).