Development of machine learning prediction model for AKI after craniotomy and evacuation of hematoma in craniocerebral trauma.

The aim of this study was to develop a machine-learning prediction model for AKI after craniotomy and evacuation of hematoma in craniocerebral trauma. We included patients who underwent craniotomy and evacuation of hematoma due to traumatic brain injury in our hospital from January 2015 to December 2020. Ten machine learning methods were selected to model prediction, including XGBoost, Logistic Regression, Light GBM, Random Forest, AdaBoost, GaussianNB, ComplementNB, Support Vector Machines, and KNeighbors. We totally included 710 patients. 497 patients were used for the training of the machine learning models and the remaining patients were used to test the performance of the models. In the validation cohort, the AdaBoost model got the highest area under the receiver operating characteristic curve (AUC) (0.909; 95% CI, 0.849-0.970) compared with other models. The AdaBoost model showed an AUC of 0.909 (95% CI, 0.849-0.970) in the validation cohort. Although there was an underestimated acute kidney injury risk for the model in the calibration curve, there was a net benefit for the AdaBoost model in the decision curve. Our machine learning model was evaluated to have a good performance in the validation cohorts and could be a useful tool in the clinical practice.