Development of a Machine Learning-Based Prediction Model for Vancomycin Steady-State Trough Concentrations

Objective: Vancomycin is a first-line glycopeptide antibiotic for infections caused by methicillin-resistant Staphylococcus aureus (MRSA), but it has a narrow therapeutic window and substantial inter-individual variability. This study aimed to develop machine-learning models for predicting steady-state vancomycin trough concentrations and to support individualized dosing in clinical practice. Methods: In this retrospective study, 206 hospitalized patient records were included, and demographic, biochemical, and other clinical data were collected. Eight machine-learning models were developed and compared: ordinary least squares (OLS) regression, least absolute shrinkage and selection operator (LASSO), ridge regression, random forest (RF), XGBoost, support vector regression (SVR), multilayer perceptron (MLP), and a stacking ensemble. Model performance was evaluated using the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and the proportions of predictions within ±20% (P20) and ±30% (P30) of observed values. SHAP was used for interpretability analysis. Results: The random forest model achieved the best performance, with an R2 of 0.751, RMSE of 4.77 mg/L, MAE of 3.48 mg/L, and P30 of 71.0% in the test set. The accuracy of three-class clinical classification was 74.2%, with no severe misclassification. Conclusion: Tree-based ensemble machine-learning methods effectively predicted steady-state vancomycin trough concentrations. A bedside calculator was developed to assist clinical dosing decisions.