Optimizing Predictive Performance of Bayesian Forecasting for Vancomycin Concentration in Intensive Care Patients

Tingjie Guo,Armand R J Girbes,Rob J Bosman,Reinier M Van Hest,Johan G C Van Hasselt,Ron A A Mathot,Laura B Zwep,Luca F Roggeveen,Paul W G Elbers,Peter H J Van Der Voort,Lucas M Fleuren

doi:10.1007/s11095-020-02908-7

Abstract

PurposeBayesian forecasting is crucial for model-based dose optimization based on therapeutic drug monitoring (TDM) data of vancomycin in intensive care (ICU) patients. We aimed to evaluate the performance of Bayesian forecasting using maximum a posteriori (MAP) estimation for model-based TDM.MethodsWe used a vancomycin TDM data set (n = 408 patients). We compared standard MAP-based Bayesian forecasting with two alternative approaches: (i) adaptive MAP which handles data over multiple iterations, and (ii) weighted MAP which weights the likelihood contribution of data. We evaluated the percentage error (PE) for seven scenarios including historical TDM data from the preceding day up to seven days.ResultsThe mean of median PEs of all scenarios for the standard MAP, adaptive MAP and weighted MAP method were − 7.7%, −4.5% and − 6.7%. The adaptive MAP also showed the narrowest inter-quartile range of PE. In addition, regardless of MAP method, including historical TDM data further in the past will increase prediction errors.ConclusionsThe proposed adaptive MAP method outperforms standard MAP in predictive performance and may be considered for improvement of model-based dose optimization. The inclusion of historical data beyond either one day (standard MAP and weighted MAP) or two days (adaptive MAP) reduces predictive performance.

Highlights

Therapeutic drug monitoring (TDM) concerns the measurement of drug concentrations in patients to optimize dosing schedules in individual patients and is commonly used in treatment optimization of patients of intensive care unit (ICU)
The creatinine clearance of the patients were calculated by default based on the MDRD formula, provided that the commonly seen formulae for calculating glomerular filtration rate were expected to perform in ICU patients [8,9]
The median percentage error (PE) of including preceding one up to seven days of historical data for the standard maximum a posteriori (MAP) method ranged from −3.4% to −11.5% and for the adaptive MAP method, −3.4% to −5.4% (Fig. 3a, b)

Summary

Introduction

Therapeutic drug monitoring (TDM) concerns the measurement of drug concentrations in patients to optimize dosing schedules in individual patients and is commonly used in treatment optimization of patients of intensive care unit (ICU). Pharm Res (2020) 37: 171 estimated using a previously developed population PK model based on collected historical drug concentration TDM data of a patient who has received the drug. These individual PK parameters in turn can be used to perform Bayesian forecasting to predict the prospective concentrations to further derive a dosing schedule that meets therapeutic concentration targets associated with efficacy or toxicity. In ICU patients it is unclear if and how such historical TDM data should be included for Bayesian forecasting, and, if current MAP-based approaches optimally make use of available historical data in estimating individual PK parameters. Vancomycin represents a relevant paradigm drug to study the optimization of model-based TDM approaches

Objectives

Methods

Results

Conclusion