Machine Learning for Mortality Prediction in Infective Endocarditis: A Systematic Review and Meta-Analysis

Infective endocarditis (IE) continues to be an often fatal condition despite improvements in cardiac surgical procedures and antibiotic therapy, and conventional scoring tools show poor generalizability. Machine learning (ML) addresses these limitations by capturing complex, nonlinear clinical relationships, outperforming conventional scores in predictive accuracy, though prior ML work in IE has focused on diagnosis. A PRISMA-compliant systematic review and meta-analysis of PubMed (Supplemental Digital Content, https://links.lww.com/CIR/A251) and Scopus (through April 2026) evaluated supervised ML models predicting all-cause mortality in adult IE patients; study quality and reporting were appraised using PROBAST and TRIPOD. Eight studies (5503 patients, mean age 53.85) were included in qualitative synthesis, of which 5 contributed area under the receiver operating characteristic curve (AUC) or C-index estimates for pooling via random-effects models, stratified into in-hospital/early and 6-month mortality subgroups. Seven studies were retrospective, and 1 was prospective. ML models, especially ensemble approaches such as Random Forest and gradient boosting, demonstrated strong discriminative performance across all cohorts, with AUC reflecting the ability to distinguish patients who died from those who survived, and outperformed conventional models. Pooled AUC was 0.85 (95% confidence interval [CI], 0.81-0.89) for in-hospital/early mortality (I2 = 35.3%) and 0.85 (95% CI, 0.82-0.88) for 6-month mortality (I2 = 0%). Dominant predictor domains varied by clinical context; multisystem physiologic markers characterized general IE cohorts, dynamic and laboratory variables enhanced intensive care unit-based predictions, and procedural and anatomical factors defined surgical and transcatheter aortic valve replacement model performance. Risk of bias was identified in 4 studies. ML models showed strong discriminative performance for IE mortality prediction, with ensemble methods outperforming conventional approaches by capturing its multivariate heterogeneity. Clinical adoption remains limited; future efforts should prioritize multicenter prospective validation, longitudinal data integration, and development of interpretable frameworks for bedside adoption.