Dynamic HRV Monitoring and Machine Learning Predict NYHA Improvements in Acute Heart Failure Patients

Abstract

Heart failure (HF) is marked by significant morbidity, mortality, and readmission rates, highlighting a critical need for accurate assessment of treatment efficacy. The New York Heart Association (NYHA) classification, while standard, falls short in capturing treatment responses. Heart rate variability (HRV), a sensitive autonomic function indicator, offers a non-invasive HF prognosis monitoring tool. This study aimed to explore dynamic changes in HRV parameters (ΔHRV) between admission and discharge as novel biomarkers for acute-to-stable phase transition in HF, leveraging wearable devices and machine learning to enhance treatment efficacy assessment.

We monitored HRV in 40 HF patients at admission and discharge using wearable devices. Statistical analysis and machine learning models were applied to assess the association between ΔHRV and NYHA classification improvements. Significant correlations were found between ΔHRV in SDNN and SD2 and NYHA enhancements (p < 0.001), with the Random Forest model achieving the highest predictive accuracy (AUC = 0.77).

This study demonstrates that ΔHRV, particularly SDNN and SD2, serves as a sensitive and non-invasive biomarker for real-time monitoring of HF treatment responses. The integration of wearable HRV monitoring with machine learning enables personalized HF management, with a focus on identifying and prioritizing high-risk patients for early intervention, thereby reducing readmission rates.