Predicting Fatal Coronary Heart Disease with ECG-AI
Author Information
Author(s): Butler Liam, Ivanov Alexander, Celik Turgay, Karabayir Ibrahim, Chinthala Lokesh, Tootooni Mohammad S., Jaeger Byron C., Patterson Luke T., Doerr Adam J., McManus David D., Davis Robert L., Herrington David, Akbilgic Oguz, Sawchuk Alan P.
Primary Institution: Wake Forest University School of Medicine
Hypothesis
Can ECG-AI models accurately predict the risk of fatal coronary heart disease (FCHD)?
Conclusion
The study found that the 2-year risk of FCHD can be predicted with high accuracy using single-lead ECGs combined with demographic information.
Supporting Evidence
- The best model achieved an AUC of 0.91 for 2-year FCHD risk prediction.
- Using Lead I ECGs alone resulted in an AUC of 0.86 for predicting FCHD risk.
- Models combining demographic data with ECG-AI predictions showed improved accuracy.
- High sensitivity (74%) and specificity (83%) were reported for the best model.
- Correlation between 12-lead and single-lead ECG-AI models was strong (R = 0.74).
- FCHD events were predicted with high accuracy using only demographic data and ECG-AI.
- Statistical significance was found in the differences between model performances across cohorts.
- Subgroup analyses showed no significant differences in model performance across sex and race.
Takeaway
Doctors can use simple heart tests to tell if someone might have a serious heart problem in the next two years.
Methodology
The study used retrospective data from ECGs and demographic information to develop and validate ECG-AI models for predicting FCHD risk.
Potential Biases
Potential biases may arise from the patient population being primarily older and with existing health issues.
Limitations
The study's results may not apply to younger populations or those without existing health conditions, and the data was collected from patients already receiving care.
Participant Demographics
The UTHSC cohort included 50,132 patients with a mean age of 62.5 years, while the AHWFB cohort included 2,305 patients with a mean age of 63.0 years.
Statistical Information
P-Value
<0.001
Confidence Interval
0.85–0.89
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website