Evaluating the Incremental Value of AI-Derived ECG Features Beyond Automated ECG Reports and Clinical Data for 30-Day Mortality: A MIMIC-IV Cohort Study

AI-derived ECG features improved 30-day all-cause mortality risk ranking in 25,430 held-out MIMIC-IV patients, reaching AUROC 0.812 versus 0.737 for ECG morphology.

• Why it matters: Automated ECG reports and clinical records already predict mortality, so it is unclear whether AI adds incremental value, generalizes across settings, and changes which patients are prioritized under limited review.
• What they did: The study compared 15 prespecified ECG morphology measurements with D-BETA features in the same held-out patients, using AUROC and review-capacity simulations at 5% and 10%, plus tests in the Brazilian CODE-15% cohort.
• The result: Incremental gains were +0.075 (95% CI +0.059 to +0.091) over morphology and +0.053 (+0.039 to +0.067) with basic clinical information, but were near zero versus the expanded automated ECG report (+0.0037, 95% CI −0.0053 to +0.0126). At 5% review capacity, AI captured more subsequent deaths (236 vs 163; 262 vs 221 after adding basic clinical data), while selective AI routing missed 47 deaths.

AI summary of the abstract · may contain errors · not medical advice · read the paper.