1Meenakshi Medical College Hospital & Research Institute, Meenakshi Academy of Higher Education and Research
2Department of Research, Meenakshi Academy of Higher Education and Research
3Department of Biochemistry, Meenakshi Ammal Dental College and Hospital, Meenakshi Academy of Higher Education and Research
4Arulmigu Meenakshi College of Nursing, Meenakshi Academy of Higher Education and Research
5Meenakshi College of Nursing, Meenakshi Academy of Higher Education and Research
6Meenakshi College of Arts & Science, Meenakshi Academy of Higher Education and Research
Aim: To design and test machine-learning models to forecast significant complications in patients with advanced heart failure that receive left ventricular assist device (LVAD) implantation.
Background: The LVAD treatment has become a necessary alternative in the patients with end-stage heart failure, but the issues that occur after the procedure, including the pump thrombosis, gastrointestinal bleeding, infection and right-ventricular failure, have become frequent and not easily predicted with the help of the traditional risk chart. Machine-learning can provide better personalized risk estimates using complicated clinical, laboratory, and hemodynamic information.
Methods: A retrospective of 1,480 LVAD patients of three developed heart-failure units were reviewed. Five algorithms: random forest, gradient boosting, support vector machine, logistic regression, and neural networks were trained using twenty-three preoperative variables, such as demographics, comorbidities, laboratory biomarkers, finding of right-heart catheterization, and measures of echocardiography. The models were evaluated with a cross-validation and tested on external validation set. One-year risk of pump thrombosis, major bleeding, and right-ventricular failure was considered the primary outcomes.
Results: Gradient boosting showed the best score in composite complications (AUC 0.86), high predictive power of right-ventricular failure (AUC 0.88). The importance of bilirubin, RA pressure, INR, RV function, and inflammatory markers were found to be the most important predictors tailored by feature-importance analysis.
Conclusion: The machine-learning models have shown a lot of promise of the early detection of the high-risk LVAD patients, enabling better decision-making and individualized perioperative care.
Index terms: Advanced heart failure, Machine learning, left ventricular assist device, predictive modeling, Right ventricular failure
How to cite this article: Suvaithenamudhan S, Purushothaman I, Anandhi D, Vasanthapriya J, Jayabharathi B, Shanthi V. Machine Learning for Prediction of Left Ventricular Assist Device (LVAD) Complications in Advanced Heart Failure. Int J Drug Deliv Technol. 2026;16(10s): 183-190; DOI: 10.25258/ijddt.16.10s.27
Source of support: Nil.
Conflict of interest: None