External validation of artificial intelligence for detection of heart failure with preserved ejection fraction.

Summary

In a matched cohort (n=496), an AI echocardiography model (EchoGo Heart Failure v2) showed similar discrimination to H2FPEF/HFA-PEFF but reduced intermediate classifications and improved net classification and clinical decision impact. Positive AI classification identified patients with a two-fold higher risk of death or HF hospitalization, adding prognostic value.

Key Findings

• AI HFpEF model achieved similar discrimination and calibration to H2FPEF/HFA-PEFF but with fewer intermediate classifications.
• Integration of AI with existing scores improved correct management decisions.
• AI-positive patients had approximately two-fold higher risk of mortality or HF hospitalization.

Clinical Implications

Integrating AI-based echocardiography analysis into HFpEF pathways can reduce indeterminate cases, prioritize further testing, and identify patients at higher risk for closer management.

Limitations

• Single-model evaluation; generalizability to diverse scanners and populations requires further study
• Observational design; no randomized clinical impact assessment

Future Directions

Prospective multi-center impact studies, integration with workflows, and calibration across devices/vendors to ensure equitable performance.