Individualized Meropenem Dosing in CRRT Patients: Development and Validation of a Machine Learning-Based Decision Support Tool

Background The optimization of meropenem dosing in critically ill patients undergoing continuous renal replacement therapy (CRRT) remains a significant clinical challenge. This study aimed to assess whether a machine learning–based clinical decision support system (CDSS) can support clinicians in determining the optimal meropenem dosing strategy. Methods Using a previously published population pharmacokinetic (POPPK) model of meropenem, we generated a virtual cohort of 2000 CRRT patients to simulate various dosing strategies and pathogen susceptibility profiles. Patient demographics, dosing parameters, and CRRT settings were incorporated as features to train ML models predicting the probability of achieving two pharmacodynamic (PD) endpoints (100% fT > MIC and 100% fT > 4×MIC). Feature importance was further assessed using SHAP analysis. External validation was conducted with clinical CRRT patient data, and predictive performance was compared against the traditional POPPK model. The top-performing model was deployed into a Streamlit-based web CDSS, named "MerDose," for clinical use. Results Across the simulated test dataset, ML models demonstrated outstanding predictive performance for both PD targets, with most accuracy and F1 scores surpassing 0.95. SHAP analysis indicated that MIC, dosing interval, administered dose, CLCRRT, and CRCL were the key determinants of predictive performance. In external validation, the Adaboost model achieved accuracies of 0.813 and 0.835 for the 100% fT > MIC and 100% fT > 4×MIC endpoints, respectively, markedly superior to the POPPK model (0.629 and 0.714). The Streamlit-based "MerDose" tool enables real-time individualized dose predictions and target attainment probabilities, offering practical guidance for initial dosing decisions in clinical care. Conclusion We developed an ML-driven CDSS, "MerDose," capable of accurately predicting individualized meropenem dosing in critically ill CRRT patients. Its web-based deployment ensures clinical applicability, offering a pragmatic decision-support framework to optimize antibiotic therapy in this high-risk population.