Model evaluation
The external evaluation of the population PK models revealed that neither of the models assessed fitted our data. The model 2 was suboptimal when describing observed concentrations in our adult CF population when only covariates remained in the dataset. Furthermore, simulation based diagnostics revealed misspecification, especially for NPDEs, which were not normally distributed. Bias and inaccuracy of model 1 were compliant with the set criteria. Even if MDAPE exceed 20 % when only covariates remained in the model, indicating that residuals and peaks at Day3 are valuable data for the Bayesian forecasting. However, simulations-based diagnostic plots were not adequate to produce relevant Monte-Carlo simulations.
Several hypotheses can be put forward to explain the inability of the two models to fit our data. (i) Demographic and clinical differences in our populations and those used in evaluated models may have a significant impact on the results. Crass et al. used a mixed CF population composed of pediatric and adult patients, which can lead to bias in the estimated parameters. (ii) Also, height was set as covariate on clearance and V1 for both investigated models. Even if height-based doses should be considered for patients with CF according to Alghanem et al. , further studies are required to understand the clinical relevance of this result (18). A weight-based tobramycin model for CF patients should have been investigated. (iii) The wide inter- and intra-individual variability in our population, especially for doses (ranging from 2.7 to 14.2 mg/kg) and dosing schedules (q12h, q24h, q36h, q48h and q72h) could explain the deviations observed in our results. The pcVPC, theoretically corrects this deviation by taking into account the dosing regimens of patients. Moreover, NPDEs further correct this discrepancy, as each individual tobramycin concentration is compared to the simulated one, taking individual dosing and covariates into account. (iv) An unknown source of variability may not have been considered. It is known that the concomitant administration of a BL affects the PK of tobramycin (19). Set as covariate, the administration of a BL may explain this inter-individual discrepancy.
A new model must be developed for this adult CF population in order to make relevant Monte-Carlo simulations.