Meta-Analysis of the Performance of AI-Driven ECG Interpretation in the Diagnosis of Valvular Heart Diseases

Comprehensive Summary

Valvular heart disease, the condition of heart valve failure, is common and often difficult to detect early because symptoms and exam findings can be unreliable. Echocardiography is the diagnostic standard, but requires resources that may not be available in all settings. Singh and his colleagues performed a meta-analysis to review ten studies that used artificial intelligence to interpret electrocardiograms for valvular heart disease detection. The studies included data from 713,537 patients. Most models used convolutional neural networks. The pooled accuracy of AI-based ECG interpretation was 81%. The combined sensitivity was 83% and the specificity was 72%. The pooled positive predictive value was low at 13%, while the pooled negative predictive value was high at 99%. Subgroup analyses for aortic stenosis and mitral regurgitation showed similar patterns with high sensitivity and negative predictive value but low positive predictive value. Ultimately, the meta-analysis covered the models' high performance with people who do not have the disease, yet weak performance for confirming positive cases.

Outcomes and Implications

Clinicians can use AI-based ECG tools to screen large patient groups efficiently and identify those who do not require further testing. The very high negative predictive value means that a normal AI-ECG result can provide reassurance and may reduce unnecessary echocardiograms. However, the low positive predictive value means that an abnormal AI-ECG result should not be considered a primary diagnostic. These findings support using AI-ECG as an initial screening step, followed by confirmatory imaging when indicated. Further work is needed to evaluate real-world implementation, validation across diverse population groups, and the effect of such screening on clinical outcomes.