Prediction of Acute Kidney Injury in Critically ill Patients with Community-Acquired Pneumonia Using Machine Learning

Background: The incidence of acute kidney injury (AKI) is increased in patients with community-acquired pneumonia (CAP), contributing to poor outcomes in ICUs. Early identification of patients at high risk for AKI is essential for timely intervention. This study aimed to develop a machine learning model for predicting AKI in CAP patients. Methods: Patients with CAP were identified from the MIMIC-IV database using ICD codes. AKI was defined according to the KDIGO criteria. Baseline characteristics, vital signs, laboratory data, comorbidities, and clinical scores were extracted. LASSO regression was applied for feature selection, and eight machine learning models, including logistic regression, k-nearest neighbors, decision tree, random forest, support vector machine, neural network, XGBoost, and LightGBM, were developed. Model performance was evaluated using AUC, sensitivity, specificity, accuracy, recall, F1 score, calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used to interpret the final model. A web-based risk calculator was created for clinical application. Results: A total of 3213 CAP patients were included, with 2723 (84.8%) developing AKI. XGBoost demonstrated the best performance with an AUC of 0.937 (95% CI: 0.922-0.952), sensitivity of 0.875, specificity of 0.855, accuracy of 0.865 (95% CI: 0.841-0.887), recall of 0.875, and F1 score of 0.866. DCA showed the highest net benefit for XGBoost across various risk thresholds. After recursive feature elimination, a simplified model with seven key variables, including urine output, weight, ventilation, first-day minimum PTT, first-day maximum sodium, first-day minimum heart rate, and first-day maximum temperature, maintained high predictive performance (AUC = 0.925, 95% CI: 0.908-0.941). Conclusions: The XGBoost model accurately predicted AKI risk in CAP patients, demonstrating robust performance and clinical utility. The web-based calculator offers an accessible tool for individualized risk assessment, supporting early detection and management of AKI in ICUs.