Artificial intelligence-enhanced electrocardiography for predicting mortality and major adverse cardiovascular events: A systematic review and narrative synthesis.
Elhaitham Yasir Awad Musa, Amar Yasir Awad Musa
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and narrative synthesis evaluated AI-enabled 12-lead or single-lead ECG models for predicting all-cause mortality, cardiovascular mortality, and major adverse cardiovascular events in adults. Thirteen studies, covering approximately 5.9 million patients, were included; most models used convolutional neural networks applied to raw ECG waveforms. For mortality prediction, reported C-statistics ranged from 0.70 to 0.89. Outcome definitions and follow-up periods varied, limiting comparisons across studies. The authors conclude that AI-ECG has prognostic potential but requires prospective validation and standardization before clinical implementation.
Key findings
- Thirteen studies included approximately 5.9 million patients; eleven assessed all-cause mortality, four cardiovascular mortality, and eight MACE.
- Reported C-statistics for mortality prediction ranged from 0.70 to 0.89.
- Most models used convolutional neural networks to analyse raw ECG waveforms.
- Outcome definitions and follow-up durations varied, limiting direct comparison between studies.
Why it matters
ECGs are widely used, and AI methods may extract prognostic information from their waveforms for mortality and cardiovascular-event risk assessment. The review suggests this approach could be scalable, but the variation in outcomes and follow-up makes it difficult to judge comparative performance or how results translate across settings. Further validation is needed to establish clinical utility.
What to keep in mind
The review reports substantial variation in outcome definitions and follow-up durations, limiting direct comparison. Its narrative synthesis reflects heterogeneity across the included studies.