Comparative Evaluation of 11 Machine Learning-Based Risk Prediction Models for Vancomycin-Associated Acute Kidney Injury During Initial Dosing

Vancomycin-associated acute kidney injury (AKI) can develop early during therapy and remains an important clinical concern. Although area under concentration-time curve (AUC)-guided dosing is recommended, the clinical relevance of early non-steady-state exposure for predicting AKI remains unclear. In this study, the authors aimed to develop and compare machine-learning models for predicting vancomycin-associated AKI using early exposure metrics.

Early AUC values, trough concentrations, concomitant medications, and comorbidities were incorporated into machine learning models. Eleven algorithms were evaluated using nested cross validation to minimize optimistic bias and prevent data leakage.

Among 442 patients (median age, 72 years), 5.7% developed AKI. Across the models, day 2 AUC and trough concentration consistently emerged as the most influential predictors, followed by concomitant use of piperacillin/tazobactam, diuretics, and vasopressors. The random forest model demonstrated the best overall discriminative performance, with an area under the receiver-operating characteristic curve of 0.89 and area under the precision-recall curve of 0.47. SHapley Additive exPlanations indicated nonlinear increases in predicted AKI risk at a day 2 AUC of approximately ≥550 mg·h/L and trough concentration of approximately ≥20 mcg/mL. These findings suggest that early vancomycin exposure, particularly cumulative exposure during the first 48 hours, is strongly associated with AKI risk.

Machine learning models integrating early AUC values and trough concentrations may provide a clinically useful framework for individualized prediction of AKI risk during the early phase of vancomycin therapy.