Mortality Risk Prediction in Patients After Cardiac Surgery Using Machine Learning

Abstract

Norwood surgery restores systemic flow in neonates with solitary ventricle inherited heart abnormalities, but it is complicated and deadly. We created machinelearning(ML) strategies to anticipate cardiac surgery mortality using preoperative and intraoperative risk indicators. Mortality after surgery was the main consequence. Sensitivity, specificity, accuracy, and AUC were used to estimate model performance. The visualization algorithm used Shapley's additive explanations. Random-forest, neural-network, SVM, and gradient-boosting had AUC-values of 0.87, 0.79, 0.81, and 0.82 for anticipating cardiac surgery operative mortality, while EuroSCORE I and II, STS, and LR had 0.70, 0.73, 0.71, and 0.74. Shapley's additive explanations study of RF identified the top-20 predictors and individual-level explanations. ML strategies based on clinical data may anticpate heart surgery postoperative mortality better than clinical grading methods. Explanatory models can account for affecting factors to provide tailored risk profiles. These machine learning-driven models need more prospective multicenter studies to prove their therapeutic efficacy.

Ähnliche Arbeiten