Voriconazole Exacerbates Subtherapeutic Linezolid Exposure in Critically Ill Patients with Staphylococcus aureus infection: Drug-Drug Interaction Evaluation and Dose Optimization Based on Population Modeling and Ensemble Learning.

OBJECTIVE: This study systematically evaluated the pharmacokinetic interactions between linezolid and voriconazole in critically ill patients with Staphylococcus aureus infections, aiming to quantify linezolid underexposure risks and construct an interpretable machine learning (ML) model for precision dosing. METHODS: A single-center retrospective study involving 48 critically ill patients undergoing therapeutic drug monitoring was conducted. A population pharmacokinetic (PPK) model was established to assess the impact of voriconazole on linezolid clearance. Monte Carlo simulations (MCS) were used to evaluate the probability of achieving efficacy and safety targets across various dosing regimens. Finally, nine base and two ensemble ML models were trained on an MCS-expanded virtual cohort using stepwise feature engineering, and were interpreted via SHapley Additive exPlanations (SHAP). RESULTS: The PPK model identified voriconazole co-administration as a key covariate significantly increasing linezolid clearance, with MCS confirming a markedly elevated risk of linezolid underexposure. Among the ML models predicting the 24-hour area under the curve, stepwise feature engineering enhanced accuracy. The Voting ensemble performed best with basic clinical features, whereas the Stacking regression model demonstrated superior predictive capability when integrating deeper mechanistic parameters, accurately revealing non-linear feature interactions via SHAP. CONCLUSION: Co-administering voriconazole in critically ill patients significantly accelerates linezolid clearance, increasing treatment failure risks at higher minimum inhibitory concentrations. This study provides a quantified dose optimization strategy and a robust, mechanism-driven ensemble ML model to support individualized clinical dosing in complex drug-interaction scenarios.