Using Machine Learning to Predict Acute Kidney Injury After Aortic Arch Surgery

Abstract

Objectives: Machine learning models were compared with traditional logistic regression with regard to predicting kidney outcomes after aortic arch surgery.

Design: Retrospective review.

Setting: Single quaternary care center, Fuwai Hospital, Beijing, China.

Participants: The study comprised 897 consecutive patients who underwent aortic arch surgery from January 2013 to May 2017. Three machine learning methods were compared with logistic regression with regard to the prediction of acute kidney injury (AKI) after aortic arch surgery. Perioperative characteristics, including patients' baseline medical condition and intraoperative data, were analyzed. The performance of the models was assessed using the area under the receiver operating characteristic curve.

Measurements and main results: The primary endpoint, postoperative AKI, was defined using the Kidney Disease: Improving Global Outcomes criteria. During the first 7 postoperative days, AKI was observed in 652 patients (72.6%), and stage 2 or 3 AKI developed in 283 patients (31.5%). Gradient boosting had the best discriminative ability for the prediction of all stages of AKI in both the binary classification and the multiclass classification (area under the receiver operating characteristic curve 0.8 and 0.71, respectively) compared with logistic regression, support vector machine, and random forest methods.

Conclusion: Machine learning methods were found to predict AKI after aortic arch surgery significantly better than traditional logistic regression.

Keywords: acute kidney injury; aortic arch surgery; cardiopulmonary bypass; machine learning.