Related Experiment Videos
Prediction of Meropenem Target Attainment in Critically Ill Patients Using Routine Clinical Data from a Single
Qingliu Lu1, Yinglian Wang2, Ke Huang1
1Department of Pharmacy, Guangxi Academy of Medical Sciences and the People's Hospital of Guangxi Zhuang Autonomous Region, Nanning, Guangxi, 530021, People's Republic of China.
Objective:
Meropenem pharmacokinetics vary considerably among critically ill patients, resulting in uncertainty in target concentration attainment. This study aimed to develop a predictive model using real-world therapeutic drug monitoring (TDM) data to assess target attainment and provide a reference for individualized meropenem management.
Methods:
This single-center retrospective study included critically ill patients administered meropenem treatment and subjected to trough concentration monitoring from August 2023 to January 2026. Meropenem target attainment was defined as a trough concentration (Cmin) exceeding 8.0 mg/L. The analysis included 399 TDM samples from 333 patients, which were partitioned into training and test sets at an 8:2 ratio using a patient-level splitting approach. Following feature selection, eight machine learning algorithms were applied to build prediction models. Predictive performance was assessed primarily by the area under the receiver operating characteristic curve (AUC), with accuracy, sensitivity, specificity, F1 score, and calibration curves used as additional evaluation measures. SHapley Additive exPlanations (SHAP) analysis was then used to interpret the final model.
Results:
Ten features were ultimately selected, including age, CRRT, ECMO, Scr, ALP, PLT, P-LCR, APACHE II score, CrCl, and dose per body weight. In the independent test set, the random forest showed the best performance, with an AUC of 0.873 (95% CI: 0.779-0.946), followed by extreme gradient boosting with an AUC of 0.867 (95% CI: 0.763-0.942) and support vector machine with an AUC of 0.863 (95% CI: 0.771-0.935). SHAP analysis showed that CrCl, APACHE II score, and Scr were the most important predictors.
Conclusion:
Machine learning models developed using routine clinical variables showed good performance in predicting meropenem target attainment in critically ill patients and may help identify patients at high risk of target non-attainment, providing a reference for individualized meropenem management.