PROVAR+
2026
AI-Enabled Tele-ECG Identifies Structural Heart Disease and Prioritizes Echocardiography in Underserved Populations: The PROVAR+ External Validation Study
In adults undergoing public-system cardiovascular screening in two Brazilian cities, an AI-enabled tele-ECG model showed better discrimination for structural heart disease on echocardiography compared with a model of clinical variables and conventional ECG abnormalities.
Risultato
Endpoint primario: Detection of structural heart disease on echocardiography
Of 3,779 participants (median age 66 years; 56.7% women), 58.7% screened AI-ECG-SHD-positive and 24.2% had structural heart disease on echocardiography. Discrimination was AUC 0.80 (95% CI 0.78-0.82) with AI-ECG-SHD vs 0.75 (95% CI 0.74-0.77) with the conventional model; p<0.001. At the predefined threshold AI-ECG-SHD achieved 59.4% accuracy, 87.3% sensitivity, 50.4% specificity, 92.6% negative predictive value, diagnostic odds ratio 7.00 and NNS 4.7. Under fixed echocardiography capacity, AI-guided prioritisation reduced the estimated time to diagnosis of 50% of cases from 407 to 172 days.
- Sottotema
- Implementation & new uses
- Disegno
- External validation of an AI-enabled tele-ECG for identifying structural heart disease and prioritising echocardiography, within PROVAR+, the Brazilian (Minas Gerais) programme of handheld echocardiographic screening in primary care
- Arruolati
- 3779 pazienti
- Fonte
- Borges e Reges R, Vinhal WC, et al. Am J Cardiol. 2026; published online 8 November.doi 10.1016/j.amjcard.2026.10.022
Discussione e critica
In a large prospective cohort where every participant had echocardiography, the AI tele-ECG model found structural heart disease better than a conventional clinical and ECG model, with high sensitivity and negative predictive value. Specificity was only 50.4% and 58.7% of participants screened positive, so the model would still send many people without disease to echocardiography. The shorter time to diagnosis comes from a simulation under fixed capacity, not from a prospective triage pathway, and no patient outcomes were measured. The cohort came from two Brazilian cities, and the abstract does not report positive predictive value, calibration or how the threshold was chosen.