Related Experiment Videos
Comparison Between Artificial Intelligence-Based Models and Traditional Risk Scores for Predicting Risks in Adult
Tobias K Fuchs1, Cameron Jones1, Michael Breiner2
1Edward Via College of Osteopathic Medicine (VCOM), Clinical Sciences, Blacksburg, Virginia.
The Journal of Surgical Research
|June 17, 2026
Summary
Machine learning (ML) models show promise in improving cardiac surgery risk prediction compared to traditional scores. Further validation is needed to address limitations like data leakage and overfitting before widespread clinical use.
Area of Science:
- Cardiovascular Surgery
- Medical Informatics
- Artificial Intelligence in Medicine
Background:
- Traditional risk scores (e.g., EuroSCORE II) may not capture complex interactions in cardiac surgery.
- Machine learning (ML) offers nonlinear modeling to potentially improve risk prediction.
Purpose of the Study:
- Systematically review studies comparing ML models with traditional risk scores for perioperative mortality and adverse events in adult cardiac surgery.
- Focus on literature published between 2020 and 2026.
Main Methods:
- Systematic review following PRISMA 2020 guidelines.
- Searched PubMed, Google Scholar, and Cochrane (Jan 2020 - Jan 2026).
- Assessed bias using Prediction model Risk Of Bias ASsessment Tool+ artificial intelligence.
Main Results:
- Thirteen studies compared ML and traditional scores.
- ML models demonstrated improved or comparable performance (AUC differences 0.006-0.42) and calibration.
- Specific algorithms like extreme gradient boosting and random forest showed high AUC for various outcomes.
Conclusions:
- ML models, particularly ensembles, may enhance cardiac surgery risk prediction.
- Further prospective multicenter validation is crucial.
- Address data leakage, overfitting, temporal drift, and algorithmic fairness for clinical integration.