Rank order of candidates for heart transplantation in France: An explainable machine learning analysis

The French composite cardiac allocation score is based on an algorithm that includes interacting components. This multistep system may hinder understanding of the decision-making process leading to candidate ranking. This study aimed to provide insights into the variables contributing to the ranking of the French transplant candidates. This national cohort study included all candidates ranked between January 2018 and December 2023. The primary endpoint was the normalized ranking of candidates. After determining the most effective machine learning model for predicting ranking, we applied Shapley additive explanations (SHAP) analysis to highlight the relationship between contributing factors and predicted ranking. During the study period, 365 310 rankings were generated, of which 197 684 rankings from 2430 donors and 3221 candidates were other than 0. The posttransplant survival filter was activated in 7% of cases. The top 5 factors contributing to the predicted ranking were the combined veno-arterial extracorporeal membrane oxygenation/natriuretic peptide variable, total bilirubin level, transport time from procurement to transplant center, candidate age, and estimated glomerular filtration rate. The candidate's sex, blood type, and indication for transplantation had minimal impact on ranking. These results show that our allocation system combines urgency and efficiency while limiting the influence of variables associated with reduced access to transplant.