Abstract 18204: Application of Machine Learning in Predicting All-Cause Mortality in Patients With Left Ventricular Assist Device
Vien T Truong, Binh Nguyen, Alex Xing X Wang, Phan Dai, Vivek Patel, Mansoor Ahmad, Muhammad Saad Siddique, Syed Muhammad Usama, Avilash Mondal, Kevin Quach, Durgesh Agrawal, Syed Fahad Shah, Sonela Skenderi, Thomas Metkus, Sunil Dhar, Eugene Chung- Physiology (medical)
- Cardiology and Cardiovascular Medicine
Introduction: Machine learning has been demonstrated to significantly enhance the accuracy of cardiovascular risk prediction across various cardiac diseases.
Hypothesis: This study sought to develop a machine learning (ML) framework to predict all-cause mortality in patients with left ventricular assist device (LVAD).
Methods: All patients in INTERMACS registry who received an LVAD from June 2006 to December 2017 were screened. The Light Gradient Boosting Machine (LightGBM) algorithm was used to train an ML model for predicting all-cause mortality. Nested ten-fold cross-validation was performed to optimize the model and estimate its performance on unseen data. Model performance was assessed using the receiver operating characteristics (ROC) curves.
Results: The final study group consisted of 22,137 patients, with a mean age of 56±13 years, and 21.3% being female. After a median follow-up of 12.4 months (IQR, 4.7 - 27.1 months). The overall all-cause mortality was reported in 7228 patients (32.7%). The proposed LightGBM-based framework had good performance with an area under the receiver operating characteristic curve (AUC-ROC) of 0.74 ± 0.05 on average over the 10 trials. Furthermore, when stratified by gender, the LightGBM-based framework had good performance (AUC-ROC of 0.73 for males and AUC-ROC of 0.70 for females).
Conclusions: A ML model using LightGBM can predict all-cause mortality in LVAD patients with good performance.