Abstract Body (Do not enter title and authors here): Background
Coronary Artery Bypass Graft Surgery (CABG) is the most commonly performed operation in cardiac surgery and results from isolated CABG are used as benchmark to rate cardiac surgery programs in the US by the Society of Thoracic Surgeons (STS). Accurate and reliable mortality risk prediction of CABG patients is essential for developing targeted treatment strategies. Traditional risk scores such as the STS score and EuroSCORE II offer moderate discriminative value, and have limited utility in predicting outcomes for high-risk patients. Machine learning (ML) models have emerged as an alternate method that may offer improved risk assessment. This study aims to compare machine learning models with traditional risk scores for predicting all-cause mortality in patients undergoing CABG.

Methods
PubMed, EMBASE, Web of Science and Cochrane databases were searched until 18^th May 2024 for studies comparing ML models with traditional statistical methods for event prediction of CABG patients. The primary outcome was comparative discrimination measured by C-statistics with 95% confidence intervals between ML models and traditional methods in estimating the risk of all-cause mortality. A secondary aim was to compare model calibration between ML models and traditional risk scores, adhering to guidelines for predictive algorithm comparisons.

Results
A total of 27 studies were included (568,190 patients). The summary C-statistic of all ML models across all endpoints was 0.82 (95% CI, 0.78-0.85), compared to traditional methods 0.73 (95% CI, 0.70-0.75). The difference in C-statistic between all ML models and traditional methods was 0.09 (p<0.0001). One model undertook external validation, and calibration was inconsistently reported.

Conclusion
ML models demonstrated superior discrimination of all-cause mortality for CABG patients compared to traditional risk scores. Whilst there is great potential for ML models to be integrated into electronic healthcare systems to improve pre-operative risk stratification, and guide clinical decision making, the methodological and validation limitations pose a hurdle for immediate clinical implementation.