When guideline-based transthoracic echocardiography cannot determine diastolic function grade in a substantial share of acute heart failure patients, artificial-intelligence-enhanced electrocardiography may be able to assign a grade. A study of 11,513 acute heart failure patients found that artificial-intelligence-enhanced electrocardiography diastolic function grades correlated with established severity biomarkers and independently predicted mortality across key subgroups.
The research, conducted by Yee Weng Wong and colleagues, was chosen as a 2025 Arthur E. Weyman Young Investigator’s Award Competition Finalist. Their findings were presented at the American Society of Echocardiography (ASE) 2025 annual meeting and as an abstract in the Journal of the American Society of Echocardiography.
Using a convolutional neural network trained on digital ECGs, researchers categorized patients into three diastolic function groups based on a single ECG. AI–enhanced electrocardiography (AI-ECG) classified 30% as normal/grade 1; 37% as grade 2; and 33% as grade 3. By comparison, guideline-based diastolic function grading by transthoracic echocardiography (TTE) was indeterminate in 44% of cases.
The analysis included adults aged 18 years or older from the Mayo Clinic Unified Data Platform with documented heart failure who received at least one dose of intravenous loop diuretic between December 2013 and December 2023, and who had paired ECG and TTE data available. Patients involved in the original AI-ECG model development were excluded to ensure independent validation, noted Wong, of the Mayo Clinic, Rochester, MN, and colleagues.
Higher AI-ECG grades tracked with greater disease severity. NT-proBNP levels rose stepwise with grade, and pulmonary capillary wedge pressure (PCWP) by right-heart catheterization was higher in grade 3 than in normal/grade 1 when available.
In adjusted Cox models, AI-ECG grades independently predicted all-cause mortality:
-
Overall: aHR 1.25 (grade 2) and 1.44 (Ggrade 3) vs normal/grade 1
-
LVEF over 40%: aHR 1.26 (grade 2) and 1.5 (grade 3)
-
LVEF 40% or less: aHR 1.28 (grade 2) and 1.44 (grade 3)
Findings were consistent in sinus and non-sinus rhythm and in patients with TTE-determinate and TTE-indeterminate DF. Kaplan–Meier curves showed clear separation by AI-ECG grade.
This was a retrospective, single-institution analysis within the Mayo Clinic Health System, which may limit generalizability. While AI-ECG can provide a DF grade when TTE is indeterminate, further work is needed to define its integration into clinical workflows.
Disclosures were not available at press time.
