Machine learning-based prediction model of acute kidney injury in patients with acute respiratory distress syndrome

Abstract

Abstract Background Acute kidney injury (AKI) can make cases of acute respiratory distress syndrome (ARDS) more complex, and the combination of the two can significantly worsen the prognosis. Our objective, therefore, is to utilize machine learning techniques to construct models that can promptly identify the risk of AKI in ARDS patients, and provide guidance for early intervention and treatment, ultimately leading to improved prognosis. Method We obtained data regarding ARDS patients from the Medical Information Mart for Intensive Care III (MIMIC-III) database and utilized 11 machine learning (ML) algorithms to construct our predictive models. We selected the best model based on various metrics, and visualized the importance of its features using Shapley additive explanations (SHAP). We then created a more concise model using fewer variables, and optimized it using hyperparameter optimization (HPO). Additionally, we developed a web-based calculator to facilitate clinical usage. Result A total of 928 ARDS patients were included in the analysis, of whom 179 (19.3%) developed AKI during hospitalization. A total of 43 features were used to build the model. Among all models, XGBoost performed the best. We used the top 10 features to build a compact model with an area under the curve (AUC) of 0.838, which improved to an AUC of 0.848 after the HPO. Conclusion Machine learning algorithms, especially XGBoost, are reliable tools for predicting AKI in ARDS patients. The compact model still retains excellent predictive ability, and the web-based calculator makes clinical usage more convenient.