In the earlier diagnostic phase, the same team showed that NAVI-HF performed well against lung ultrasound as a reference standard. That phase established that the device, which analyzes heart and lung sounds with machine learning algorithms, could identify congestion patterns consistent with those seen on imaging. The new prognostic phase asked a different and clinically important question: does a positive NAVI-HF result actually predict worse outcomes?

To answer it, the researchers enrolled 239 patients being discharged after treatment for acute heart failure. Before discharge, each patient underwent NAVI-HF assessment and was classified as either positive or negative for congestion. Of the cohort, 111 patients, or 45.1 percent, were NAVI-HF positive.

The investigators then followed all participants for a median of six months, tracking a composite outcome that captured adverse clinical events during that window. The baseline characteristics told a consistent story: NAVI-HF positive patients had a worse clinical profile and higher NT-proBNP levels, with differences reaching statistical significance at p less than 0.05. During follow-up, 82 composite events occurred, representing 34.3 percent of the cohort.

In univariate Cox regression analysis, a positive NAVI-HF result was associated with a significantly increased risk of the composite outcome, with a hazard ratio of 1.59 and a 95 percent confidence interval of 1.03 to 2.47, corresponding to a p-value of 0.03. The device’s status also predicted heart failure hospitalization specifically, with a hazard ratio of 1.64 and a confidence interval of 1.05 to 2.46, at a p-value of 0.02. All-cause mortality, however, was not significantly associated with NAVI-HF status, with a p-value of 0.24.

The pattern of associations shifted when the researchers adjusted for other variables. The significant links between NAVI-HF positivity and both the composite outcome and heart failure hospitalization persisted after adjustment for age and sex. But when the analysis was extended to include other prognostic variables, the associations were attenuated and lost statistical significance.

Whether NAVI-HF adds independent predictive value beyond those markers remains an open question that the present data cannot definitively resolve. The present study supports the idea that the congestion signal is prognostically meaningful, while also tempering expectations about independence from existing risk measures.

Several limitations frame the interpretation. The attenuation of the associations after multivariable adjustment means the device should be viewed as a congestion detector whose prognostic value is partly redundant with established measures rather than as a wholly independent predictor. The authors phrase the central conclusion carefully, as an association between NAVI-HF–detected congestion and adverse outcomes rather than a demonstrated causal pathway or a proven clinical benefit.

For now, the message from this cohort of 239 patients is measured: what the device hears before discharge is not meaningless sound, but a signal that, in unadjusted and partially adjusted analyses, tracked with the patients’ six-month clinical course.

Subject of Research: Prognostic value of an AI-based auscultation device for detecting congestion in patients discharged after acute heart failure

Article Title: Clinical outcome implications of novel auscultation device of artificial intelligence (NAVI-HF) in predischarge acute heart failure patients

Article References: Santoso, R. M., Wee, S., Huang, W., Yeo, Y. K., Soesanto, A. M., Jatmiko, W., Kekalih, A., Indrasari, N. D., & Siswanto, B. B. B. (2026). Clinical outcome implications of novel auscultation device of artificial intelligence (NAVI-HF) in predischarge acute heart failure patients. Frontiers in Cardiovascular Medicine, 13, 1864004. https://doi.org/10.3389/fcvm.2026.1864004

Image Credits: AI Generated

DOI: 10.3389/fcvm.2026.1864004

Keywords: artificial intelligence, heart failure, auscultation, lung ultrasound, prognosis, NT-proBNP, pulmonary congestion, machine learning, hospitalization, cohort study, cardiology, medical devices