A Machine Learning Method for Predicting Acute Kidney Injury in Patients with Intracranial Hemorrhage.

Abstract

Intracranial hemorrhage (ICH) is a critical and urgent condition in clinical practice. Recent research has highlighted acute kidney injury (AKI) that frequently impacts patient prognosis. For clinicians, early intervention is crucial, and the advancement of machine learning brings promising prospects for predicting this disease. Therefore, this study aims to develop innovative machine learning models for the prediction and diagnosis of acute kidney injury (AKI) in patients with intracerebral hemorrhage (ICH). AKI data of ICH patients were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. To construct the models, we utilized various techniques including random survival forest (RSF), elastic network (Enet), Least Absolute Shrinkage and Selection Operator (Lasso), stepwise logistic regression (stepwise LR), and ten machine learning algorithms. Optimal parameters were obtained through a ten-fold crossover, and training and testing groups were employed for the integrated machine models' training and validation. We conducted a quantitative evaluation of the model's performance and assessed its clinical application to determine its advantages. Furthermore, we compared the base model with traditional models such as the Sequential Organ Failure Assessment (SOFA) and the bespoke Simplified Acute Physiology Score (SAPS) II model. A total of 1856 patients with intracerebral hemorrhage (ICH) were enrolled in the study, consisting of 1633 non-AKI patients and 223 AKI patients. Among the various machine learning models tested, XGBoost exhibited the highest predictive accuracy and demonstrated excellent clinical applicability as a standalone model. When combining integrated models, RSF+XGBoost, LR[forward]+Lasso, LR[forward]+RSF, and Lasso+XGBoost, all achieved the highest AUC values (AUC = 1.000). Machine learning models can serve as valuable diagnostic tools in identifying the occurrence of acute kidney injury (AKI) in intracerebral hemorrhage (ICH) cases. Whether utilized individually or in combination, these models have the potential to assist clinicians in proactively developing effective interventions.