An exploratory prospective clinical study of AI-based voice analysis for predicting heart failure severity

Heart failure (HF) remains a major global health burden, affecting over 64 million individuals worldwide. Accurate assessment of HF severity is pivotal for clinical management but typically relies on resource-intensive methods such as imaging and laboratory biomarkers. Voice analysis represents a novel, scalable, and non-invasive approach to monitor physiological changes linked to HF severity.

This study aims to develop and validate an artificial intelligence (AI)-based voice analysis model to predict HF severity in patients undergoing treatment and in control subjects.

This prospective clinical study is being conducted across multiple academic hospitals. We will recruit HF patients receiving treatment and matched control subjects without HF. Voice recordings will be collected at multiple time points: at discharge, and at 1, 5, and 13 weeks post-discharge. Simultaneously, clinical data including NT-proBNP levels, NYHA functional class, dyspnea scale scores, and echocardiographic findings will be gathered. An AI model will be trained on extracted acoustic features and clinical variables to classify HF severity into four categories: normal, mild, moderate, and severe.

We hypothesize that changes in voice characteristics will reflect dynamic shifts in pulmonary and systemic congestion, correlating with traditional HF severity markers. The AI model’s performance will be evaluated using standard metrics including area under the curve (AUC), sensitivity, and specificity, and compared against established clinical benchmarks.

This study is expected to demonstrate the feasibility of voice-based AI tools for non-invasive monitoring of HF severity. Such technology may enable continuous, remote assessment of patients, enhancing clinical decision-making and patient management in heart failure care.