Korean Artificial Intelligence-Based Risk Model for Predicting Mortality After Acute Myocardial Infarction

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Predictive Analytics

Background:

• Acute myocardial infarction (AMI) carries significant long-term mortality risk.
• Existing risk models may not reflect outcomes with contemporary treatments.
• Need for updated, accurate risk prediction tools in AMI management.

Purpose of the Study:

• To develop and validate simplified machine learning models for predicting 3-month and 3-year mortality after AMI.
• To assess the performance of these models in a contemporary therapeutic setting.
• To provide a reliable tool for identifying high-risk AMI patients.

Main Methods:

• Development of the Korean Artificial Intelligence-Based Risk Model for Acute Myocardial Infarction (KARMA) using a boosted decision tree algorithm.
• Utilized the Korea Acute Myocardial Infarction Registry (KAMIR)-NIH registry (2011-2015) for model development and internal validation.
• External validation performed using the KAMIR-V registry (2016-2020) with seven routinely available clinical variables.

Main Results:

• KARMA models showed excellent discrimination for 3-month (AUC 0.91) and 3-year (AUC 0.85) mortality.
• KARMA significantly outperformed GRACE and KAMIR scores in predictive accuracy.
• Consistent performance across subgroups and robust external validation confirmed reliability.

Conclusions:

• KARMA scores offer a reliable and accessible method for predicting post-AMI mortality.
• Facilitates identification of high-risk individuals for optimized management.
• Aims to improve patient outcomes through enhanced risk stratification.