Predicting Waitlist Mortality for Liver Transplant Candidates: A Comparative Analysis between Statistical Scores and Machine Learning Models

Accurately predicting waitlist mortality for liver transplant candidates is a critical yet challenging task. Traditional models such as MELD, MELD-Na, and MELD 3.0 have been widely used by clinicians but fall short in delivering precise mortality predictions when compared to machine learning (ML) models. In this study, we conduct a comprehensive comparative analysis of these conventional scoring systems against advanced ML models, including LDA, TabNet, Random Forest, and LightGBM. Results not only highlight the improved predictive accuracy of certain ML models over MELD-based scores but also identify the most significant variables influencing 3-month waitlist mortality. This analysis enables the proposal of new, critical risk factors for consideration in future scoring models. By leveraging these insights, we aim to contribute to the development of a more efficient and equitable organ allocation system, ultimately enhancing patient outcomes and potentially saving more lives through better patient prioritization.