Machine learning approaches for the prediction of 1-year mortality after heart transplantation: A systematic review and meta-analysis

Abstract

AIM: We aimed to compare the discrimination between machine learning (ML)-based models and conventional risk scores for predicting 1-year all-cause mortality after heart transplantation. METHODS: This study was conducted in accordance with PRISMA and TRIPOD-AI guidelines. PubMed, EMBASE, Web of Science, and Cochrane were searched to February 2026 for studies directly comparing ML models with conventional risk scores in adult heart transplant recipients. The primary outcome was comparative discrimination for 1-year all-cause mortality, quantified using C-statistics with 95% confidence intervals (CIs). Random-effects meta-analysis pooled discrimination estimates and assessed between-group differences. Prespecified subgroup analyses were performed according to validation status and ML subtype. RESULTS: Eight studies including 14 models and 305,186 recipients were analysed (Figure). Summary C-statistic for 1-year mortality was higher for ML than conventional scores (0.68 [0.66-0.70] vs 0.62 [0.60-0.64]; absolute difference 0.06 [0.03-0.08]; p<0.001). Random-forest ensembles were the top performing model on subgroup analysis (0.67 [0.65-0.69]). External validation was reported in 37.5% of models, with inconsistent calibration. Across all SHAP analysis of all models, age, creatinine, and bilirubin were strongest predictors of mortality. CONCLUSIONS: ML models demonstrated modestly higher discrimination for 1-year mortality after heart transplantation; however, limited external validation and inconsistent calibration restrict immediate clinical applicability.