Evaluate the predictive performance of amikacin plasma concentration using the Bayesian approach in critically ill patients at Bach Mai Hospital

Purpose: This study aimed to externally validate the population pharmacokinetic (popPK) model of amikacin and assess the applicability of Bayesian forecasting approach in individualizing amikacin dosing for critically ill patients. Materials and Methods: Retrospective data from 121 ICU patients with 231 blood amikacin concentrations were used to evaluate the predictive performance of the model published by A.H. Nguyen which was developed previously from patients in the same department. The model was evaluated based on a priori and a posteriori approach using relative bias (rBias), relative root mean squared error (rRMSE), and Visual Predictive Checks. For Bayesian forecasting, the study investigated the ability to predict peak concentrations by using solely a middle level in dosing interval. The rBias and rRMSE and the clinical agreement evaluation based on a threshold of 45 mg/L was applied for this purpose. The forecast performance between Bayesian estimation and first-order pharmacokinetic calculation was compared based on trough concentrations, with Bias and RMSE indices. Results: The A.H. Nguyen model was appropriate for both a priori and a posteriori prediction, with rBias values of 18.40% and 2.42%, respectively. The Bayesian forecasting method demonstrated good predictive performance when comparing forecasted peak concentrations to observations, with rBias of 6.81% and rRMSE of approximately 30%. An excellent agreement (over 85%) between observed- and forecast-peak concentrations was achieved. No difference was found in forecasting trough concentration by Bayesian estimation using one-level or two-level. However, the trough level estimated by the Bayesian methods resulted in a positive bias of 1.25 mg/L in comparison with first-order PK calculation. Conclusions: The A.H. Nguyen model could be applicable in a model-based TDM for Vietnamese ICU patients. Bayesian forecasting using only middle concentration might be sufficient for the dose individualization of amikacin for both efficacy and safety monitoring.