An ECG-based machine-learning approach for mortality risk assessment in a large European population

Area of Science:

• Cardiology
• Machine Learning
• Public Health

Background:

• Electrocardiograms (ECGs) are standard diagnostic tools in cardiology.
• Predicting long-term mortality risk can aid in patient management and preventative strategies.

Purpose of the Study:

• To assess the 5-year all-cause mortality risk in a European population using a machine learning approach.
• To evaluate the predictive performance of electrocardiogram (ECG) parameters, age, and sex for mortality risk.

Main Methods:

• A cohort of 53,692 patients aged 40-90 years with ECG recordings was analyzed.
• Logistic regression models were applied to predict mortality, excluding patients with severe ECG abnormalities.
• Model performance was assessed using metrics like the area under the receiver operating characteristic curve (AUC).

Main Results:

• Over 26% of patients died within 5 years.
• The logistic regression model showed significant predictive capability across age groups, with an average AUC of 0.779.
• The model effectively distinguished between survivors and non-survivors based on predicted mortality probability.

Conclusions:

• The developed machine learning model demonstrated good predictive performance for all-cause mortality.
• This highlights the potential of ECGs as a prognostic tool beyond their diagnostic applications.
• ECG parameters, combined with age and sex, can effectively predict long-term mortality risk in specific patient populations.