Prediction of Mortality Using Machine Learning Algorithms in Post-CABG Patients with ECMO Support

Background: Patients with acute myocardial infarction (AMI) undergoing emergency coronary artery bypass grafting (CABG) with postoperatively extracorporal membrane oxygenation (ECMO) support are associated with higher mortality and morbidity. To identify factors leading to poor outcome in these patients is of significant importance. In this study we predicted the mortality of these patients through machine learning algorithms considering the significant risk factors.

Method: In this retrospective study, we evaluated 100 patients in our hospital with AMI undergoing CABG between 2008 and 2017 with ECMO support. A comprehensive amount of patients data (~200 variables) in hourly diuresis of the patients in the 1st 48 hours after the operation, Encourage score, APACHE II score, EuroSCORE II, treatment with catecholamines in the first 7 days, fluid balance in the first 7 days, duration of ECMO treatment, daily evaluation of organ function and many more demographic and clinical data were used to develop different machine learning models. We used multiple state-of-the-art machine learning models to predict mortality in these Patients. The machine learning models used included decision trees, multilayer perceptron, random forest, AdaBoost, support vector machine, logistic regression, and Bernoulli naive Bayes model. We refrained from oversampling. The training-to-test set ratio was 70:30. The test set area under the curve was used as metric to compare model performance.

Results: Our results indicate that the best performing models to predict mortality in CABG patients with ECMO support are based on random forest and Bernoulli naive Bayes. The best classification performance was 0.83 for naive Bayes and 0.82 for random forest. The computing time of the trained models was no longer than 1.5 seconds per patient.

Conclusion: Identifying risk factors in Patients with AMI with low output syndrome (LOS) and ECMO support in an early stage is of significant importance to reduce mortality and morbidity. Our results indicate that models based on Bernoulli naive Bayes and random forest are most suited for mortality prediction in AMI patients undergoing CABG surgery with ECMO support. A prospective validation in a separate cohort with large sample size is recommended to support our study.

Publication History

Article published online:
28 January 2023