Abstract Introduction Teicoplanin is commonly used to treat Gram-positive bacterial infections in the intensive care unit (ICU). However, evidence to support individualized therapeutic drug monitoring (TDM)-guided daily dosing of teicoplanin in pediatric ICU patients remains limited despite substantial interpatient variability in pharmacokinetics and clinical response. Aim To develop and validate a real-world TDM-informed machine learning model to predict physician-adjusted teicoplanin daily dose in pediatric ICU patients, with the goal of supporting individualized dosing decisions in clinical pharmacy practice. Method Clinical and TDM data from pediatric ICU patients receiving teicoplanin at the Sun Yat-sen Memorial Hospital of Sun Yat-sen University between June 2020 and June 2023 were retrospectively collected. The outcome variable was the daily teicoplanin dose administered during routine TDM-guided clinical care. After univariate screening and sequential forward selection, the dataset was divided into training and test sets (8:2). Missing values were imputed using the random forest approach. Nine machine learning and deep learning algorithms, including gradient boosting, XGBoost, LightGBM, and TabNet, were developed and evaluated using tenfold cross-validation, with model performance assessed using the coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute error (MAE). Results A total of 257 pediatric ICU patients (595 teicoplanin dosing records) were included in the study. Weight, age, height, teicoplanin trough concentration (TDM), glucose, creatine kinase isoenzyme-MB, total protein, concomitant imipenem and meropenem use, and upper respiratory infection were identified as key predictors. Among the nine models, the TabNet algorithm demonstrated the best performance on the test set (R 2 = 0.82, RMSE = 53.96 mg/day, MAE = 39.93 mg/day). The proportion of predictions within ± 30% of the observed daily dose was 81.51%. Conclusion This real-world TDM-informed TabNet model shows strong performance in predicting the daily dose of clinician-adjusted teicoplanin in pediatric ICU patients. The model may serve as a clinical decision-support tool for pharmacists and physicians to assist individualized teicoplanin dosing within routine TDM workflows, potentially improving dosing consistency, and supporting safe and effective antimicrobial therapy.
Wang et al. (Fri,) studied this question.