Implementation of novel auscultation device of artificial intelligence (NAVI-HF) in predischarge acute heart failure patients

Introduction Residual pulmonary congestion detected by lung ultrasound (LUS) occurs in approximately 40% of predischarge acute heart failure (HF) patients and is associated with worse outcomes. Purpose This study evaluated the diagnostic performance and prognostic significance of a novel auscultation device of artificial intelligence (NAVI-HF) with LUS as reference standard to detect pulmonary congestion in predischarge acute HF patients. Materials and Methods This multicenter prospective cohort study enrolled predischarge acute HF patients with reduced ejection fraction. The first phase evaluated diagnostic performance of NAVI-HF, while the second phase assessed prognostic significance. NAVI-HF recorded thoracic acoustic signals at five predefined chest locations, and a support vector machine classifier categorized patients as NAVI-HF positive or negative based on presence of B-lines on LUS. Patients were followed for six months for composite outcomes of all-cause mortality and HF hospitalization. Results A total of 221 patients were included (mean age 57.5 years; 76.5% male), with median LVEF of 32% and NT-proBNP level of 3272 pg/mL. LUS identified pulmonary congestion in 53.4% of patients, whereas conventional auscultation detected crackles in 26.7%. Using LUS B- lines pulmonary congestion findings as the reference standard, NAVI-HF demonstrated sensitivity, specificity, accuracy, PPV, NPV, and F1 score of 99%, 96%, 98%, 97%, 99%, and 0.98 in the training dataset. In the testing dataset, the corresponding values were 91%, 82%, 86%, 75%, 94%, and 0.82. NAVI-HF positive was associated with higher NT-proBNP, lower LVEF, and greater comorbidities burden of hypertension, diabetes, and chronic kidney disease with worse NYHA class (all p<0.05). After a median follow-up of 6 months, 68 (30.8%) composite outcome occurred. Univariate cox regression analysis showed that NAVI-HF positive was associated with a significantly increased risk of the composite outcome (HR 1.98; 95% CI 1.21–3.23) and HF hospitalization (HR 2.10; 95% CI 1.26–3.46), but not all-cause mortality (p = 0.24). These associations for composite and HF hospitalizations outcome remained significant after adjustment for age and sex (p 0.03 and p 0.02), yet were attenuated and no longer significant after further adjustment for established prognostic variables (p 0.60 and p 0.50). Conclusion NAVI-HF demonstrated high diagnostic performance for detecting pulmonary congestion, including subclinical cases, and is associated with worse clinical outcomes.