Abstract 14978: Using a Machine Learning Risk Score to Predict Outcomes in Patients Undergoing LVAD Implantation: An Analysis of the INTERMACS Database

Abstract

Background: Left ventricular assist device (LVAD) implantation is a viable treatment option for end-stage heart failure patients. Despite its benefits, outcomes of LVAD surgery can vary significantly, underscoring the importance of accurately assessing patients' risk prior to the procedure. MARKER-HF is a mortality risk model that utilizes machine learning algorithms to predict outcomes in HF patients. Aim: We assessed the predictive accuracy of MARKER-HF in patients undergoing LVAD implantation and determined whether it offered prognostic information beyond the INTERMACS profile. Methods: We calculated the MARKER-HF score in 25,365 patients from the INTERMACS database who underwent LVAD implantation between 2006 and 2020 and divided the population into 4 risk groups based on their score (low <-0.2; moderate: -0.2 - 0; high: >0 - 0.2; very high: >0.2). The outcomes of interest were in-hospital and post-discharge 1-year mortality. Results: There was a negative correlation between the MARKER-HF score and INTERMACS profile (Pearson r = -0.348). MARKER-HF score exhibited a wide distribution within each INTERMACS profile. During the index hospitalization, 1383 (5.5%) patients died. The in-hospital mortality increased steadily with a higher MARKER-HF score, ranging from 2.9% in low to 13.8% in very high-risk groups, while it decreased with a higher INTERMACS profile. When combining the MARKER-HF and INTERMACS profile, we identified both patients with the lowest (2.6%) and the highest in-hospital mortality (16.1%). In a Cox-proportional hazard regression analysis adjusting for both MARKER-HF and INTERMACS profile, only the MARKER-HF score was significantly associated with increased post-discharge mortality (hazard ratio: 1.25, P<0.001), but not INTERMACS profile (P=0.847). Conclusion: The MARKER-HF score is a simple non-invasive tool that can assist with optimal decision-making and can add additional information beyond the INTERMACS classification.