A simulation study to assess the influence of population pharmacokinetic model selection on initial dosing recommendations of vancomycin in neonates.

Aims: The accuracy of model-informed precision dosing largely depends on selecting the most appropriate population pharmacokinetic (popPK) model from many available options. This study aims to evaluate the concordance of optimal initial simulated doses among various vancomycin popPK models developed in neonates and to explore the role of predictive performance in explaining the variability in probability of target attainment (PTA).

Methods: A virtual neonatal patient population was created and 26 previously externally evaluated vancomycin popPK models were used to simulate 5 different dosing regimens. For each simulated scenario, the area under the concentration-time curve and PTA were calculated to assess the agreement on optimal initial doses across the 26 models. A multiple regression was performed to explore the impact of the models' predictive performance on PTA.

Results: For most models (15/26), there was an agreement on the optimal dosing regimen. The highest PTA being achieved by the model with the best a priori predictive performance. The multiple regression model significantly predicted mean ln-transformed PTA, with F(2, 23) = 5.406 and P = .010, yielding an adjusted R of .21. PTA was significantly influenced by imprecision (P = .048) but not bias (P = .469).

Conclusion: In conclusion, our study demonstrated that, despite the variability in bias and imprecision, there was a consensus on the initial optimal doses for the majority of models; however, models with superior a priori predictive performance yielded higher PTA values. Bias and imprecision alone only seem to predict a small proportion of the variability in PTA, with imprecision having a more pronounced effect.