Abstract Body (Do not enter title and authors here): Background: Heart failure with reduced ejection fraction (HFrEF) remains significantly underdiagnosed in low-resource settings due to limited access to echocardiography and cardiology expertise. Artificial intelligence (AI)-enabled digital diagnostics offer transformative potential for early detection and task shifting. We conducted the first prospective validation and implementation study of an AI-powered digital stethoscope (EKO SENSORA™) for detecting left ventricular ejection fraction (LVEF) <40% in a sub-Saharan African population.
Methods: The Detection and Management of Heart Failure with SENSORA in Underserved Nations (DAMSUN-HF) study prospectively enrolled 115 adults with cardiopulmonary symptoms between May and July, 2025, using a hub-and-spoke recruitment model that integrated district hospitals and primary clinics in Ghana. A priori power calculations indicated that 100 subjects would provide >90% power to detect sensitivity ≥90%, assuming ≥50% prevalence. All participants underwent AI-assisted cardiac auscultation using the SENSORA™ platform, followed by blinded transthoracic echocardiography (TTE). The primary outcome was the diagnostic accuracy (sensitivity/specificity) of the ELEFT algorithm in detecting LVEF <40%. A parallel implementation protocol was deployed, in which non-specialist providers performed initial screenings and transmitted data to cardiologists via a secure digital platform.
Results: The mean participant age was 62 ± 16 years; 53% were female. LVEF <40% was confirmed by TTE in 59% of participants. The ELEFT algorithm achieved 97% sensitivity (95% CI: 89–99) and 80% specificity (95% CI: 67–89), with 95% negative predictive value and 86% positive predictive value. The AI-based murmur detection algorithm demonstrated 87% sensitivity and 97% specificity for systolic murmurs, and 81% and 94% respectively for diastolic murmurs. The task-shifted workflow achieved >95% adherence, and >90% of flagged cases were reviewed by a cardiologist within 48 hours. Deployment required no additional imaging, infrastructure, or disruption to existing workflows.
Conclusion: The SENSORA™ platform demonstrated high diagnostic accuracy and was successfully deployed within a decentralized, task-shifted hub-and-spoke model. These findings support broader integration of AI diagnostic tools to expand heart failure detection capacity in health systems with limited specialist access.