Prediction of risk for acute kidney injury and its progression to mortality in obese patients admitted to ICU postoperatively

‍Objective‍ ‍To develop a machine learning-based risk prediction model for postoperative acute kidney injury (AKI) and a model for mortality in obese patients admitted to intensive care unit (ICU) in order to improve early warning and prognostic evaluation to support clinical decision-making. Methods‍ ‍Data of obese postoperative ICU patients were retrospectively retrieved from the MIMIC-IV and eICU databases for statistical analysis. Ultimately, 2 520 patients (670 from MIMIC-IV and 1 850 from eICU databases) were included to build the risk prediction models for AKI and mortality. The data included demographic information, vital signs, laboratory findings, surgical types, comorbidities, and medication use. After data cleaning and preprocessing, Boruta feature selection was applied, followed by the construction of prediction models using 7 machine learning algorithms, that is, Gradient Boosting Machine (GBM), Generalized Linear Model (GLM), k-Nearest Neighbors (KNN), Naïve Bayes (NB), Neural Network (NNET), Support Vector Machine (SVM), and XGBoost. Model performance was evaluated through cross-validation and external validation. Results‍ ‍In the risk prediction models of AKI, the SVM model achieved the highest AUC value of 0.80 in the testing set and 0.71 in the external validation test. For the risk prediction models of mortality, the GBM model outperformed others in the prediction, attaining an AUC value of 0.91 in the testing set. Conclusion‍ ‍Risk predictive models for postoperative AKI and mortality in obese ICU patients are successfully constructed, and are valuable tools for clinicians to optimize early intervention and improve clinical outcomes for the patients.