Feature Importance and Dimensionality Reduction
After manual and automated variable filtering, the relative importance of each feature in predicting the primary outcome was assessed using a random forest classifier. Continuous and categorical variables were assessed both separately and together to reduce biases (Supplemental Figure 1 ). In the combined assessment, the most important features included total bilirubin, mechanical ventilation, no ventricular assist device at transplant, serum creatinine, donor age, recipient height, donor cerebrovascular mechanism of death, prior cardiac surgery, and Karnofsky functional status. The top 20 most important features from each individual assessment were then combined excluding duplicates, resulting in 47 training variables that were used for subsequent model training (Supplemental Table 1 ).