Identifying Patients for Cardiovascular Evaluation Using Artificial Intelligence and Team-Based Care in Ghana: The H-TRIAGE Study

Background: Cardiovascular diagnostic capacity is limited in many low-resource settings. We evaluated the HeartLINK Triage for Risk-Informed Allocation and Guided Evaluation (H-TRIAGE) index, a prespecified strategy combining clinical information and AI-enabled digital auscultation to prioritize cardiovascular evaluation. Methods: In this prospective study in Ghana (February-July 2026), consecutively enrolled adults underwent a standardized assessment by trained frontline nurses and community health workers. H-TRIAGE used predefined additive symptom, vital-sign, and AI-stethoscope weights to assign Low (0-3), Intermediate (4-7), or High (≥8) Risk. Four cardiologists, blinded to H-TRIAGE, independently reviewed all participants; a fifth adjudicated disagreements. The reference standard (adjudicated indication) was a recommendation for further diagnostic testing, cardiologist-led evaluation, or both. The primary cohort comprised 5 facilities across 3 districts of the Ashanti Region; external validation used a held-out facility in a separate district of the Eastern Region. In a modeled allocation analysis, Intermediate- and High-Risk participants were considered prioritized; we calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: Among 3,800 participants (primary, n=3,123; external validation, n=677; median age, 52 years [IQR, 39-64]; 77.7% women), 1,707 (44.9%) had an adjudicated indication. By H-TRIAGE category, the proportion with an indication was 2.5% (Low), 87.9% (Intermediate), and 100% (High) (P for trend<0.001). Prioritizing Intermediate- and High-Risk participants yielded 1,656 true positives, 142 false positives, 1,951 true negatives, and 51 false negatives: sensitivity, 97.0% (95% CI, 96.1%-97.8%); specificity, 93.2% (92.1%-94.3%); PPV, 92.1% (90.8%-93.3%); and NPV, 97.5% (96.7%-98.1%). This approach would prioritize 47.3% of participants for evaluation. Conclusions: H-TRIAGE accurately identified adults needing cardiologist-adjudicated evaluation, and sensitivity remained high at a geographically distinct site. Applied by frontline nurses and community health workers, it could focus scarce specialist and diagnostic capacity on the patients most likely to need it, deferring roughly half of those assessed.