Machine learning methods for surgery cancellation

Surgery cancellation has a greatly negative impact on both hospital and patient. A commonly used method for improving this problem is analyzing cancellation reasons through statistical analysis, while the improvement is inefficient. Based on surgery data from the hospital information system (HIS) of West China Hospital (WCH), we proposed to use machine learning methods include Random forests (RF), Bagging, Boosting and Bayesian additive regression trees (BART) to predict surgery cancellation. The experimental result shows that the Boosting with Over-sampling performed well in prediction accuracy (Sensitive: 0.5234, Specific: 0.8105, Accuracy: 0.7651 and AUC: 0.6670). And our research also points out the latent causes for surgery cancellation. It's useful for hospital managers.