Voice-Driven Mortality Prediction in Hospitalized Heart Failure Patients: A Machine Learning Approach Enhanced with Diagnostic Biomarkers

TL;DR

This work investigates whether voice biomarkers can predict $5$-year mortality in hospitalized heart failure (HF) patients. It develops a logistic regression model with Ridge regularization using vocal features extracted by Disvoice from CAPE-V recordings, with mutual information and LASSO-based feature selection, and enhances performance by incorporating the diagnostic biomarker NT-proBNP. The acoustic predictor shows a strong association with outcomes ($p<0.001$) and, after NT-proBNP incorporation, achieves higher accuracy than the acoustic model alone. The approach offers a non-invasive, deployable tool for risk stratification and resource allocation in HF care, while acknowledging limitations from small sample size and gender imbalance and outlining avenues for broader validation and multimodal data integration.

Abstract

Addressing heart failure (HF) as a prevalent global health concern poses difficulties in implementing innovative approaches for enhanced patient care. Predicting mortality rates in HF patients, in particular, is difficult yet critical, necessitating individualized care, proactive management, and enabling educated decision-making to enhance outcomes. Recently, the significance of voice biomarkers coupled with Machine Learning (ML) has surged, demonstrating remarkable efficacy, particularly in predicting heart failure. The synergy of voice analysis and ML algorithms provides a non-invasive and easily accessible means to evaluate patients' health. However, there is a lack of voice biomarkers for predicting mortality rates among heart failure patients with standardized speech protocols. Here, we demonstrate a powerful and effective ML model for predicting mortality rates in hospitalized HF patients through the utilization of voice biomarkers. By seamlessly integrating voice biomarkers into routine patient monitoring, this strategy has the potential to improve patient outcomes, optimize resource allocation, and advance patient-centered HF management. In this study, a Machine Learning system, specifically a logistic regression model, is trained to predict patients' 5-year mortality rates using their speech as input. The model performs admirably and consistently, as demonstrated by cross-validation and statistical approaches (p-value < 0.001). Furthermore, integrating NT-proBNP, a diagnostic biomarker in HF, improves the model's predictive accuracy substantially.

Figures (6)

• Figure 1: Collection of patient voice illustrated
• Figure 2: Interface of a website for SHFM
• Figure 3: The multi-step feature preprocessing pipeline is meticulously crafted to curate the most representative features for predicting the target label in each classification model employed in this study. Initially, the pipeline assigns reliable Mutual Information (MI) scores to each feature in the training set, followed by a subsequent step involving a backward elimination algorithm coupled with the LASSO shrinkage technique.
• Figure 4: Box plots illustrating the distributions of five features utilized in the modeling process, with respect to the prognosis target label
• Figure 5: Confusion matrix evaluated on LR performance using hold-out dataset