A machine learning model that emulates experts' decision making in vancomycin initial dose planning

Abstract

Vancomycin is a glycopeptide antibiotic that is a primary treatment for methicillin-resistant Staphylococcus aureus infections. To enhance its clinical effectiveness and prevent nephrotoxicity, therapeutic drug monitoring (TDM) of trough concentrations is recommended. Initial vancomycin dosing regimens are determined based on patient characteristics such as age, body weight, and renal function, and dosing strategies to achieve therapeutic concentration windows at initial TDM have been extensively studied. Although numerous dosing nomograms for specific populations have been developed, no comprehensive strategy exists for individually tailoring initial dosing regimens; therefore, decision making regarding initial dosing largely depends on each clinician's experience and expertise. In this study, we applied a machine-learning (ML) approach to integrate clinician knowledge into a predictive model for initial vancomycin dosing. A dataset of vancomycin initial dose plans defined by pharmacists experienced in vancomycin TDM (i.e., experts) was used to build the ML model. Although small training sets were used, we established a predictive model with a target attainment rate comparable to those of experts, another ML model, and commonly used vancomycin dosing software. Our strategy will help develop an expert-like predictive model that aids in decision making for initial vancomycin dosing, particularly in settings where dose planning consultations are unavailable.

Keywords: Initial dosing regimen; MRSA; Machine learning; TDM; Vancomycin.