Evaluation of Random Forest and Ensemble Methods at Predicting Complications Following Cardiac Surgery

Citation

Lapp L, Bouamrane M, Kavanagh K, Roper M, Young D & Schraag S (2019) Evaluation of Random Forest and Ensemble Methods at Predicting Complications Following Cardiac Surgery. In: Riano D, Wilk S & ten Teije A (eds.) Artificial Intelligence in Medicine: 17th Conference on Artificial Intelligence in Medicine, AIME 2019, Poznan, Poland, June 26–29, 2019, Proceedings. Lecture Notes in Computer Science, LNCS volume 11526. Cham: Springer International Publishing, pp. 376-385. https://doi.org/10.1007/978-3-030-21642-9_48

Abstract
Cardiac patients undergoing surgery face increased risk of postoperative complications, due to a combination of factors, including higher risk surgery, their age at time of surgery and the presence of co-morbid conditions. They will therefore require high levels of care and clinical resources throughout their perioperative journey (i.e. before, during and after surgery). Although surgical mortality rates in the UK have remained low, postoperative complications on the other hand are common and can have a significant impact on patients’ quality of life, increase hospital length of stay and healthcare costs. In this study we used and compared several machine learning methods – random forest, AdaBoost, gradient boosting model and stacking – to predict severe postoperative complications after cardiac surgery based on preoperative variables obtained from a surgical database of a large acute care hospital in Scotland. Our results show that AdaBoost has the best overall performance (AUC = 0.731), and also outperforms EuroSCORE and EuroSCORE II in other studies predicting postoperative complications. Random forest (Sensitivity = 0.852, negative predictive value = 0.923), however, and gradient boosting model (Sensitivity = 0.875 and negative predictive value = 0.920) have the best performance at predicting severe postoperative complications based on sensitivity and negative predictive value.

Keywords
Risk Prediction; Perioperative Medicine; Machine Learning; Performance Evaluation