Editor's Note
Machine learning can be used to improve surgical risk prediction compared to traditional risk calculators, this study from Singapore finds.
The traditional Combined Assessment of Risk Encountered in Surgery and the American Society of Anesthesiologists Physical Status models were compared with machine learning models in the prediction of 30-day postoperative mortality and need for ICU stay more than 24 hours in 90,785 patients at Singapore General Hospital.
Machine learning models used random forest, adaptive boosting, gradient boosting, and support vector machines. All models were evaluated on the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC).
The traditional models achieved high AUROC by predicting all negative in a predominantly negative dataset. Gradient boosting was the best performing machine learning model with AUPRCs of 0.23 for mortality and 0.38 for ICU admission outcomes.
AUPRC should be used to evaluate model predictive performance instead of AUROC when the dataset is imbalanced, the researchers say.
