Hott, Marcus

Abstract Aims Monitoring pulmonary congestion in chronic heart failure (HF) reduces decompensation and hospitalization, but conventional methods such as weight and symptom tracking are often unreliable. As fluid accumulation affects the lungs and vocal tract, subtle voice alterations may serve as a non-invasive signal for early detection of worsening HF. Methods and results The Voice Analysis for Monitoring Patients with HF trial (VAMP-HF, NCT06566911) prospectively enrolled 104 patients hospitalized with acute decompensated HF (ADHF) across two academic centres in the USA and Germany. Daily voice recordings were collected from admission to discharge, with breathing features extracted from speech and acoustic features from sustained vowels. A machine-learning model was trained to classify recordings as admission-phase vs. discharge-phase using leave-one-patient-out. Patients with clinical deterioration, insufficient audio quality, or short length of stay were excluded. Seventy-nine patients were included in the final dataset. The model classified admission and discharge with an F1-score of 0.83 (95% CI: 0.77–0.90; AUC = 0.90). In patients with higher audio volume (N = 54), performance reached 0.89 (95% CI: 0.82–0.94; AUC = 0.91). When applied to intermediate hospitalization days, model-predicted scores showed progressive increases from admission towards discharge. Performance remained robust irrespective of significant weight loss during hospitalization. Conclusion In this pilot study, structured voice and breathing analysis discriminated hospitalization phase from admission through discharge in patients with ADHF. This non-invasive approach captured progressive changes during the hospital course and warrants further investigation with concurrent objective congestion markers to establish physiological specificity.

Abstract Background Effective fluid status monitoring in chronic heart failure (HF) patients is critical for preventing decompensation and hospitalization. Conventional methods, such as daily weight tracking, often fail to detect early, pre-symptomatic fluid retention. As fluid accumulation affects the vocal tract, subtle voice alterations can serve as a biomarker for congestion. In this work, we present preliminary results from the first 55 patients enrolled at the German Heart Center at Charité (DHZC) within the multicenter VAMP-HF study, assessing the performance of AI-driven voice analysis for non-invasive fluid status monitoring. Methods The study design incorporated a run-in period for the first 20 patients to refine the recording setup and ensure sufficient audio quality for subsequent data collection. An additional three patients were excluded due to clinical deterioration or in-hospital mortality. Participants provided daily voice recordings (sustained vowels, standardized text, and varying sentences) alongside clinical measurements such as NT-proBNP, daily weight, and left-ventricular ejection fraction. A speaker-independent XGBoost classifier was trained on acoustic features to classify patient states (admission, discharge, and intermediate states). Performance was evaluated using a nested leave-one-patient-out approach, ensuring strict training and validation data separation. Results The final analysis included 32 patients (mean age 76.12 ± 12.65 years). At admission, patients had a mean NT-proBNP of 9575.55 ± 8202.87 pg/mL and a mean weight loss of 4.2 ± 4.29 kg until discharge. The trained machine learning model achieved an F1-score of 0.75 (Fig. 1b), demonstrating its ability to distinguish between admission and discharge states, based solely on voice samples. An exemplary prediction for a test patient is shown in Fig. 1a. Notably, the model was also able to detect subtle vocal changes even in patients with <2 kg weight loss. Conclusion Our findings suggest that AI-driven voice analysis is a feasible biomarker for detecting heart failure decompensation. While not directly compared with conventional early warning thresholds, these results support the hypothesis that voice-based biomarkers may serve as an early indicator of fluid congestion. Further validation in larger cohorts is needed to confirm these findings and enable integration into remote HF monitoring programs.