(W-064) Comparative evaluation of different vancomycin population pharmacokinetic models to predict Bayesian-estimated vancomycin AUC in Korean patients

Eun Kyoung Chung, Department of Pharmacy, College of Pharmacy – Pharm.D.,Ph.D, Kyung Hee University,Seoul,Republic of Korea

Objective: The objective of this study was to compare the performance of the prior model in commercial Bayesian software and a newly developed local population pharmacokinetic model for predicting concentration in the Korean hospitalized patients. In this article we emphasized the importance of selecting appropriate models used in Bayesian software for vancomycin dosing adjustment.

Methods: We reviewed EMRs for hospitalized patients treated with vancomycin to extract demographic, vancomycin concentrations, and other relevant clinical data. Previously published models and a newly developed model using our local EMR data were evaluated in Korean hospitalized patients. Predictive performance of models was assessed by Mean Abosolute Error (MAE), Root Mean Square Error (RMSE). Additionally, we evaluated dosing adjustment (AUC-based) differences resulting from selection of different models. Bayesian-estimated vancomycin AUC was predicted based on a single trough concentration (Ctr) per dosing interval using the literature models and a new population pharmacokinetic model developed by NONMEM based on vancomycin concentrations in the EMRs. The agreement in vancomycin serum concentration between the prior model and our newly developed model was assessed using Bland-Altman plots.

Results: Overall, 176 vancomycin trough concentrations measured between 0 and 30 minutes before vancomycin dosing in 125 patients were extracted from the EMRs and included in the analysis. 66 vancomycin trough concentrations were included in external validation. There were significant differences in the predictive performance. The local population pharmacokinetic model (i.e., KHNMC population model) showed the best predictive performance compared to previously published models. The KHNMC model showed the least MAE and RMSE for IPRED. (KHNMC model: MAE 10.49, RMSE 13.78, Goti model: MAE 14.60, RMSE 22.27, Thomson: MAE 13.70, RMSE 20.94) The KHNMC model predicted significantly larger Bayesian-estimated AUC than the literature models (Local model: 851.22 ± 332.0 mg·h/L, Goti model: 614.63 ± 259.4 mg·h/L , Thomson model: 604.42 ± 318.42 mg·h/L, P < 0.05), more frequently requiring vancomycin dose reduction.

Conclusions: Bayesian-estimated AUCs based on vancomycin Ctr are highly variable depending on the selected prior model for a specific patient population. Overall accuracy and precision should be considered to determine the most appropriate model for Bayesian software in specific patient. Compared to previously published models, the local KHNMC model demonstrated better predictive accuracy; this could be attributed to a patient population differences between the model building population and external validation population. Local population pharmacokinetic models might more accurately reflect the target patient population characteristics.

Citations: [1] Thomson AH, Staatz CE, Tobin CM, Gall M, Lovering AM. Development and evaluation of vancomycin dosage guidelines designed to achieve new target concentrations. J Antimicrob Chemother. 2009;63(5):1050-1057. doi:10.1093/jac/dkp085 [2] Goti V, Chaturvedula A, Fossler MJ, Mok S, Jacob JT. Hospitalized Patients With and Without Hemodialysis Have Markedly Different Vancomycin Pharmacokinetics: A Population Pharmacokinetic Model-Based Analysis [published correction appears in Ther Drug Monit. 2019 Aug;41(4):549]. Ther Drug Monit. 2018;40(2):212-221.