AI-derived score to predict effort intolerance and adverse outcome across the heart failure spectrum

Abstract

Background: Peak oxygen consumption (VO₂) is a prognostic indicator in heart failure (HF), but cardiopulmonary exercise testing (CPET) has limited feasibility. Purpose: To develop an AI-driven model to predict effort intolerance (i.e., VO₂<16 mL/kg/min) in patients across HF spectrum. Methods: The model was derived in a cohort of 1,333 subjects-351 with reduced (<50%, HFrEF), 371 with preserved (>50%, HFpEF) left ventricular ejection fraction (LVEF), 611 with cardiovascular risk factors or structural heart disease without HF (Stages A-B)-and externally validated in a cohort of 1,101 subjects. All participants underwent laboratory test, rest echocardiography, CPET. Results: A neural network including age, sex, body mass index (BMI), haemoglobin, systolic mitral annulus tissue velocity (S'), systolic pulmonary artery pressure (sPAP), and β-blocker therapy achieved good discrimination (AUC 0.86±0.01 in derivation; 0.76±0.06 in validation). A simplified AI-VO₂ score (BMI, haemoglobin, LV S', sPAP) showed good performance (AUC 0.79±0.05) and predicted HF hospitalization or all-cause death (adjusted HR 1.06 per point; 95% CI 1.03-1.10) in the derivation cohort. External validation confirmed the performance (AUC 0.73±0.02; unadjusted HR 1.18 per point, 95% CI 1.13-1.24). Conclusion: The AI-VO₂ score could identify patients with effort intolerance and adverse outcomes in HF spectrum.