Title: Evaluating Time-to-Event Models for Predicting Waitlist Mortality in Heart Transplantation via an Updated Longitudinal UNOS Dataset

Abstract:

The allocation of hearts to candidates on the transplant waitlist is currently governed by medical committees that weigh various clinical factors, yet this evaluation process remains largely informal and subjective. As the United Network for Organ Sharing (UNOS) has accumulated extensive longitudinal data on donors, organs, and patients since 2018, there is a growing demand for analytical tools to assist clinicians in making timely decisions when organs become available. This research benchmarks machine learning algorithms that utilize longitudinal waitlist histories to perform time-dependent, time-to-event modeling of mortality risk for waitlisted patients. Utilizing a dataset comprising 23,807 patient records and 77 distinct variables, we assessed model performance regarding survival prediction and discrimination capabilities over a one-year period. The top-performing model demonstrated superior accuracy compared to earlier iterations, achieving a C-Index of 0.94 and an AUROC of 0.89. The analysis identified key predictors consistent with established risk factors, while also uncovering new associations. These results offer valuable insights to enhance urgency assessments and inform policy adjustments in the heart transplantation decision-making process.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC