Demographic-aware temporal graph attention for fair and accurate cardiac abnormality detection in 12-lead ECG
An ECG AI model diagnosed heart abnormalities accurately while reducing sex-based performance gaps from 15% to 2%, showing fairness and accuracy can advance together.
DA-GAT-v2 integrates demographic conditioning into a graph attention ECG classifier to simultaneously improve diagnostic accuracy (AUROC 0.9762) and reduce sex-based performance disparity from 15.4% to 1.8% in multi-label cardiac abnormality detection. The approach provides a blueprint for algorithmic fairness in ECG AI without the traditional accuracy-fairness trade-off.
What the study was
- Study design
- Retrospective AI model development with cross-dataset validation
- Population
- PTB-XL (21,507 ECG recordings) + Chapman-Shaoxing (10,646 recordings) — publicly available datasets
- Sample size
- 32153
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Sci Rep
Why it surfaced
Algorithmic fairness in ECG AI is clinically meaningful; cross-dataset validation adds strength. Single-author study on public datasets — no prospective clinical validation. Race/ethnicity not evaluated (metadata absent).
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.