A deep learning-based electrocardiogram risk score for long term cardiovascular death and disease

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• The resting electrocardiogram (ECG) is a common cardiovascular diagnostic tool.
• The extent of long-term cardiovascular risk information derivable from resting ECGs remains unclear.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) capable of predicting long-term cardiovascular risk from resting ECGs.
• To assess the performance of the developed model, SEER (Stanford Estimator of Electrocardiogram Risk), in predicting cardiovascular mortality and atherosclerotic cardiovascular disease (ASCVD).

Main Methods:

• Utilized a large dataset of resting 12-lead ECGs from Stanford University Medical Center to train a deep CNN.
• Validated SEER's predictive accuracy for 5-year cardiovascular mortality and 5-year ASCVD using independent test sets from multiple institutions.
• Evaluated SEER's ability to reclassify patient risk when used in conjunction with existing risk assessment tools like the Pooled Cohort Equations.

Main Results:

• SEER achieved an AUC of 0.83 for predicting 5-year cardiovascular mortality in a held-out test set and comparable AUCs in external validation cohorts.
• SEER predicted 5-year ASCVD with an AUC of 0.67, demonstrating modest correlation with the Pooled Cohort Equations.
• SEER reclassified 16% of patients from low to moderate risk, identifying individuals with significantly higher actual 10-year ASCVD risk.
• The model also showed predictive capability for conditions like heart failure and atrial fibrillation, even using a single ECG lead.

Conclusions:

• Deep learning models, such as SEER, can accurately predict long-term cardiovascular risk using only resting ECG data.
• SEER offers a valuable tool for improving cardiovascular risk stratification, potentially enhancing clinical decision-making and guiding preventative therapies like statin use.
• The integration of SEER with existing risk assessment tools can lead to more precise identification of high-risk individuals.