Performance of the 12-lead ECG in predicting short- and long-term risk of sudden cardiac death

Area of Science:

• Cardiology
• Medical Informatics

Background:

• Sudden cardiac death (SCD) remains a significant cause of mortality.
• Predictive models for SCD are crucial for early intervention in high-risk populations.

Purpose of the Study:

• To evaluate the performance of 12-channel electrocardiogram (ECG) parameters in predicting SCD.
• To assess prediction accuracy across different time intervals (long-term and short-term).

Main Methods:

• Retrospective analysis of 17,625 high-risk cardiac patients with coronary angiography data.
• Development and validation of extreme gradient boosting models using digital ECG parameters.
• Evaluation of model performance with and without clinical risk factors and observed ECG changes.

Main Results:

• Single ECG predictions yielded modest Area Under the Curve (AUC) values (0.59-0.68).
• Adding clinical risk factors slightly improved AUC (up to 0.71).
• Incorporating observed ECG changes during follow-up provided the best prediction performance (AUC 0.72 unbalanced, 0.66 balanced).

Conclusions:

• 12-channel ECG parameters have a modest ability to predict SCD.
• Dynamic ECG changes over time are valuable predictors of SCD in high-risk patients.
• Machine learning models integrating ECG data and clinical factors show promise for SCD risk stratification.