Comparing the Performance of Machine Learning Models and Conventional Risk Scores for Predicting Major Adverse...

Area of Science:

• Cardiology
• Medical Informatics
• Data Science

Background:

• Machine learning (ML) models show potential for enhanced clinical utility over traditional risk scores like TIMI and GRACE.
• A knowledge gap exists regarding the comparative performance of ML versus traditional models in predicting MACCEs in AMI patients undergoing PCI.

Purpose of the Study:

• To systematically review and critically appraise studies comparing ML models and conventional risk scores for MACCE prediction in AMI patients post-PCI.
• To evaluate the performance of ML algorithms against established risk scores in this specific patient population.

Main Methods:

• A systematic search of nine academic and electronic databases (Jan 2010–Dec 2024) was conducted.
• Included were 10 retrospective studies (N=89,702) of AMI patients undergoing PCI, comparing ML and conventional risk scores for MACCE prediction.
• Studies were assessed for bias using three validation tools; most had a low risk of bias.

Main Results:

• Random forest and logistic regression were common ML algorithms; GRACE and TIMI were the most used conventional scores.
• ML models demonstrated superior predictive performance (AUC: 0.88) compared to conventional scores (AUC: 0.79) for mortality risk.
• Age, systolic blood pressure, and Killip class were key predictors in both ML and conventional models.

Conclusions:

• ML-based models exhibit superior discriminatory performance for predicting MACCEs in AMI patients post-PCI compared to conventional risk scores.
• Current models often rely on nonmodifiable factors; incorporating modifiable factors (psychosocial, behavioral) is crucial.
• Further multicenter prospective studies with external validation are needed to confirm these findings and address limitations.