AUTHOR=Liu Yi-Xi , Wen Haini , Niu Wan-Jie , Li Jing-Jing , Li Zhi-Ling , Jiao Zheng TITLE=External Evaluation of Vancomycin Population Pharmacokinetic Models at Two Clinical Centers JOURNAL=Frontiers in Pharmacology VOLUME=Volume 12 - 2021 YEAR=2021 URL=https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2021.623907 DOI=10.3389/fphar.2021.623907 ISSN=1663-9812 ABSTRACT=Background Diverse vancomycin population pharmacokinetic models in neonates have been published, however, their extrapolated predictive performances remain unknown. This study aims to evaluate the external predictability and explore the factors that might influence on model performance. Methods Published population pharmacokinetic models in neonates were screened from the literature and were evaluated by datasets from two clinical centers, including 171 neonates with 319 concentrations. The predictive performance was assessed by prediction- and simulation-based diagnostics and Bayesian forecasting. Furthermore, the influence of model structure and identified covariates was also investigated. Results Eighteen published models were included after systemic literature search. In prediction-based diagnostics, no model had median prediction error (MDPE) ≤ ± 15%, median absolute prediction error (MAPE) ≤ 30%, and the percentage of prediction error that fell within ± 30% (F30) > 50%. The simulation-based visual predictive check of most models showed large deviations between the observations and simulations. After Bayesian forecasting with one and two prior observations, the predicted performance was improved significantly. Weight, age, and serum creatinine were identified as the most important covariates. Moreover, employing maturation model based on the weight and age as well as nonlinear model incorporation of serum creatinine could improve the predictive performance significantly. Conclusion The predictability of the pharmacokinetic models is closely related to the covariate modelling approach. Bayesian forecasting can significantly improve the extrapolation predictive performance.