Cor Vasa 2026, 68(5):590-599 | DOI: 10.33678/cor.2026.033

Machine Learning-Based Prediction of Long-Term Survival and Failure to Rescue after Coronary Artery Bypass Grafting: A Systematic Review

Merlinda Dwintasari^{a, b}, Danang Himawan Limanto^{a, b}, Chikita Nur Rachmi^b, Oky Revianto Sediono Pribadi^b, Yan Efrata Sembiring^{a, b}

^a Departement of Thoracic, Cardiac, and Vascular Surgery, Faculty of Medicine, Airlangga University, Surabaya, Indonesia

^b Departement of Thoracic, Cardiac, and Vascular Surgery, Dr. Soetomo General Teaching Hospital, Surabaya, Indonesia

Introduction: Coronary artery bypass grafting (CABG) remains the gold standard for surgical revasculariza tion in complex coronary artery disease. Although short-term mortality has improved substantially, long -term survival and failure to rescue (FTR) after major complications remain clinically relevant. Traditional regression-based risk models may insufficiently capture complex nonlinear interactions among perioperative variables. This review synthesizes contemporary evidence regarding determinants of survival and the role of machine learning (ML) in outcome prediction after CABG.

Methods: A PRISMA-guided systematic search of PubMed and ScienceDirect was conducted for studies pub lished between January 2015 and January 2024. Eligible studies included adult patients undergoing isolated coronary artery bypass grafting (CABG) that reported survival outcomes or failure-to-rescue (FTR) using stati stical or machine learning-based predictive models. Relevant model performance metrics and key predictive variables were extracted and qualitatively synthesized.

Results: From 3,701 identified records, 26 studies met the inclusion criteria and were included in the qualita tive synthesis. One-year survival after CABG ranged from 87-94%, while 5-year survival ranged from 67-88% and 10-year survival from 52-75%. The overall incidence of failure to rescue (FTR) was approximately 2-3%, with markedly higher mortality observed following major complications such as renal failure (~25%) and stroke (~13%). Consistent predictors of adverse outcomes included advanced age, reduced left ventricu lar ejection fraction, diabetes mellitus, renal dysfunction, and cerebrovascular disease. Machine learning models-particularly ensemble algorithms combined with SHAP interpretability techniques-demonstrated improved predictive accuracy and discriminatory performance compared with traditional regression-based approaches.

Conclusion: Long-term survival after CABG remains heterogeneous and influenced by multifactorial clinical determinants. Machine learning-based risk stratification provides improved predictive accuracy and inter pretable individualized assessment. Integration of validated artificial intelligence tools into CABG workflows may enhance perioperative decision-making and reduce failure-to-rescue rates.

Methods: A PRISMA-guided systematic search of PubMed and ScienceDirect was conducted for studies published between January 2015 and January 2024. Eligible studies included adult patients undergoing isolated coronary artery bypass grafting (CABG) that reported survival outcomes or failure-to-rescue (FTR) using statistical or machine learning-based predictive models. Relevant model performance metrics and key predictive variables were extracted and qualitatively synthesized.

Results: From 3,701 identified records, 26 studies met the inclusion criteria and were included in the qualitative synthesis. One-year survival after CABG ranged from 87-94%, while 5-year survival ranged from 67-88% and 10-year survival from 52-75%. The overall incidence of failure to rescue (FTR) was approximately 2-3%, with markedly higher mortality observed following major complications such as renal failure (~25%) and stroke (~13%). Consistent predictors of adverse outcomes included advanced age, reduced left ventricular ejection fraction, diabetes mellitus, renal dysfunction, and cerebrovascular disease. Machine learning models-particularly ensemble algorithms combined with SHAP interpretability techniques-demonstrated improved predictive accuracy and discriminatory performance compared with traditional regression-based approaches.

Conclusion: Long-term survival after CABG remains heterogeneous and influenced by multifactorial clinical determinants. Machine learning-based risk stratification provides improved predictive accuracy and interpretable individualized assessment. Integration of validated artificial intelligence tools into CABG workflows may enhance perioperative decision-making and reduce failure-to-rescue rates.

Keywords: Artificial intelligence, Coronary artery bypass grafting, Failure to rescue, Machine learning, Risk prediction, Survival

Received: March 6, 2026; Revised: March 6, 2026; Accepted: April 4, 2026; Prepublished online: October 13, 2026; Published: October 20, 2026  Show citation