Machine learning model for predicting the risk of AKI in early hemodynamically stable sepsis patients: a study based on the MIMIC IV database.

Background: The increasing prevalence and mortality of sepsis and acute kidney injury (AKI) pose significant threats to the survival of critically ill patients worldwide. Consequently, it's essential to identify patients with sepsis-associated AKI early and to ascertain its risk factors. This study developed a machine learning (ML) model to identify patients at high risk of AKI among the hemodynamically stable sepsis. Methods: This study extracted clinical data from hemodynamically stable sepsis patients in the MIMIC IV Database for analysis, and collected clinical data from a hospital for external validation. The patients were randomly divided into a training set (70%) and a testing set (30%). Potential risk factors were screened out through univariate analysis and LASSO regression. Subsequently, models were constructed using Classification and Regression Tree, Gradient Boosting Machine, Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost). The performance of these models was assessed to determine the optimal predictive model. Results: A total of 8,276 hemodynamically stable sepsis patients were included in this study, among whom 3,061 patients (37%) experienced AKI. Nine risk factors were identified for model development using 70% of the data for training. By evaluating the model performance based on indicators including AUC, accuracy, sensitivity, kappa value, MCC, F1 score, Brier score, and DCA curve, the model constructed by XGBoost demonstrated the optimal performance. The external validation was consistent with the results. Conclusion: The XGBoost model exhibited optimal performance, and is suitable for predicting the risk of AKI in early hemodynamically stable sepsis patients.