Table 3.

Summary of publications describing artificial intelligence application in predicting mechanical circulatory support outcomes.

Author	Title	Data source used	Primary outcome	Best model performance	Study conclusion
Kourou (31)	Prediction of time dependent survival in HF patients after VAD implantation using pre- and post-operative data	Local institutional data at the university of Leuven, Belgium	Time dependent survival	Artificial neural network model had an accuracy of 84.5%, sensitivity of 87%, specificity of 82%.	Application of feature selection and prediction algorithms for variable selection significantly improved prediction ability.
Ayers (28)	Predicting survival after extracorporeal membrane oxygenation by using machine learning	Local institutional data at the University of Rochester, NY	Survival	A deep neural network model had an AUC of 0.92.	Improved prediction of survival to discharge for VA-ECMO with ML versus SAVE score.
Bellavia (32)	Usefulness of regional right ventricular and right atrial strain for prediction of early and late right ventricular failure following a LVAD implant: A machine learning approach	Local data from ISMETT center and Papa Giovanni XXIII Hospital, Italy, 2010–2017	Right ventricular failure post LVAD	Naïve Bayes achieved an AUC of 0.78 in predicting Acute RVF, 0.86 in predicting Chronic RVF, and 0.92 in predicting Acute and Chronic RVF.	No single parameter is capable of predicting post LVAD RVF and the combination of risk scores, clinical, imaging and hemodynamic profiles provides the best risk assessment.
Kilic (33)	Using machine learning to improve risk prediction in durable left ventricular assist devices	(INTERMACS), 2006–2016	90-day and 1-year survival	XGBoost algorithm had an AUC of 0.74 for 90 -day mortality, and 0.714 for 1-year mortality.	Machine learning modeling had discriminatory performance, alone or as an adjunct to logistic regression.
Kilic (30)	Machine learning approaches to analyzing adverse events following durable LVAD implantation	ENDURANCE Trial (post-hoc analysis of a prospective, randomized controlled trial)	Adverse events	Hierarchical clustering was used to categorize adverse events.	Machine learning can identify distinct time patterns of post LVAD complications, facilitating research and quality improvement.
Misumi (34)	Prediction of aortic valve regurgitation after continuous-flow LVAD implantation using artificial intelligence trained on acoustic spectra	Local institutional data, Osaka University Hospital, Osaka, Japan, 2015–2017	Development of aortic insufficiency	The ensemble model had an accuracy of 0.91 and AUC of 0.73.	Machine learning trained on acoustic spectra is promising in diagnosing LVAD complications.
Shad (35)	Predicting post-operative right ventricular failure using video-based deep learning	Multicenter (3) registry, United states	Development of post LVAD RV failure	A convolutional neural network model had an AUC of 0.729, 95% CI: 0.623–0.835.	Machine learning can outperform a team of human experts.
Hendren (29)	Phenomapping a novel classification system for patients with destination therapy LVAD	INTERMACS, 2008–2017	survival, adverse events	Unsupervised machine learning clustering analysis.	Machine learning can help identify phenogroups who have differing survival and rates of adverse events post LVAD implantation.

INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support. LVAD, left ventricular assist device. strong recommendation, weak recommendation/promising but not ready to implement, as suggested by the respective paper.