Machine learning redevelopment of GRACE, ACEF, and TIMI scores for 6-month mortality

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Traditional coronary artery disease (CAD) risk scores face challenges in clinical application due to evolving pathophysiological understanding.
• Existing studies suggest machine learning (ML) algorithms outperform traditional methods in risk prediction, but heterogeneity in datasets and models limits generalizability.

Purpose of the Study:

• To evaluate the clinical performance of the GRACE, ACEF, and TIMI risk scores for predicting mortality in acute myocardial infarction (AMI) patients.
• To enhance the predictive accuracy of these scores through redevelopment using six ML algorithms.

Main Methods:

• A retrospective study of 1,682 AMI patients undergoing percutaneous coronary intervention (PCI) from January 2019 to December 2020.
• Six ML algorithms were used to redevelop the GRACE, TIMI, and ACEF risk scores.
• Model performance was assessed using accuracy, sensitivity, precision, F1-score, AUROC, and AUPRC.

Main Results:

• Original GRACE, TIMI, and ACEF scores showed AUROC values of 0.84, 0.68, and 0.69, respectively.
• Redeveloped ML models significantly improved predictive performance, with ensemble algorithms showing the best results.
• The redeveloped GRACE score achieved an AUROC of 0.89 and AUPRC of 0.32 in temporal validation, outperforming redeveloped ACEF and TIMI scores for predicting six-month mortality.

Conclusions:

• The GRACE score demonstrated superior predictive performance for six-month mortality compared to ACEF and TIMI scores.
• ML-based redevelopment substantially enhanced the predictive accuracy of traditional risk scores.
• ML algorithms and recalibration of parameter weights improve risk prediction alignment with current clinical practice.