Prediction of Acute Kidney Injury in Intracerebral Hemorrhage Patients Using Machine Learning.

Background: Acute kidney injury (AKI) is prevalent in patients with intracerebral hemorrhage (ICH) and is associated with mortality. This study aimed to verify the predictive accuracy of different machine learning algorithms for AKI in patients with ICH using a large dataset. Methods: A total of 1366 ICH patients received treatments between 2001 and 2012 from the Medical Information Mart for Intensive Care-III (MIMIC-III) database were identified based on the ICD-9 code: 431. The main outcome of AKI during hospitalizations was confirmed based on the KDIGO criteria. Overall, ICH patients were randomly divided into the training cohort and validation cohort with the ratio of 7:3. Six machine learning algorithms including extreme gradient boosting, logistic, light gradient boosting machine, random forest, adaptive boosting, support vector machine were trained in the training cohort with the 5-fold cross-validation method to predict the AKI. The predictive accuracy of those algorithms was compared by area under the receiver operating characteristics curve (AUC). Results: A total of 1213 ICH patients were included with the incidence of AKI being 29.3%. The incidence of AKI was 29.3% among the 1213 patients with ICH. The AKI group had higher 30-day mortality (p<0.001), longer ICU stay (p<0.001), and longer hospital stay (p<0.001). Among the six machine learning algorithms, the random forest performed the best in predicting AKI in both the training cohort (AUC=1.000) and the validation cohort (AUC=0.698). The top five features in the random forest algorithm-based model were platelets, serum creatinine, vancomycin, hemoglobin, and hematocrit. Conclusion: The random forest algorithm-based predictive model we developed incorporating important features, including platelet count, serum creatinine level, vancomycin level, hemoglobin level, and hematocrit level, performed the best in predicting AKI among patients with ICH.