Abstract 4340081: Machine Learning-Based Prediction of Right Ventricular Failure Following Left Ventricular Assist Device Implantation

Background: Right ventricular failure (RVF) is a significant and potentially fatal complication following left ventricular assist device (LVAD) implantation. Clinically, RVF post-LVAD is difficult to accurately predict. Machine learning (ML) offers a promising approach to predict RVF after LVAD. Objective: To develop and evaluate machine learning models for the prediction of RVF following LVAD implantation. Methods: A comprehensive set of clinical, laboratory, echocardiographic, and demographic variables was utilized to train six machine learning classification models: decision tree, logistic regression, random forest, k-nearest neighbors, support vector machine, and gradient boosting. Each model was trained over 20 iterations using a 9:1 train-test split. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC). Logistic regression weight analysis was employed to identify clinically relevant predictive variables for early and total RVF. Results: We analyzed 246 patients who underwent left ventricular assist device (LVAD) implantation at our center between January 2004 and December 2017. All patients underwent right heart catheterization (RHC) within 30 days prior to implantation and transthoracic echocardiography (TTE) within 30 days post-implantation to assess for early or late right ventricular failure (RVF). Early RVF was defined as the need for an unplanned right ventricular assist device (RVAD) within 30 days after LVAD implantation or requirement for more than 14 days of continuous inotropic support. Late RVF was defined as patients requiring medical intervention following the index hospitalization. ML models robustly predicted early RVF (AUROC: 0.769–0.841) and total RVF (0.765–0.850), with the random forest algorithm demonstrating the best performance for both. Models predicting late RVF were not as robust (0.467–0.593). Logistic regression weight analysis identified pulmonary artery pulsatility index (PAPi), global longitudinal strain (GLS), right ventricular dP/dt, and alanine aminotransferase (ALT) as clinically relevant predictors of early and total RVF. Conclusions: ML models reliably predicted early and total RVF following LVAD implantation. These findings support the potential utility of ML models in improving risk stratification to guide clinical decision-making in this high-risk population.