Development and validation of a home telemonitoring-based alert system for prediction of heart failure events using non-invasive, commercially available devices

Background/Introduction High readmission rates remain a major challenge in heart failure (HF) management. Although home telemonitoring has been investigated for early detection of HF deterioration, many validated systems rely on invasive or specialised devices, limiting their widespread clinical implementation. Clinically implementable predictive systems based solely on simple, non-invasive, commercially available devices have not been sufficiently established. Purpose We aimed to develop and validate a practical alert system capable of predicting HF events using data obtained from commercially available home monitoring devices. Methods In this prospective multicentre study, 87 patients hospitalised for acute HF or acute decompensation of chronic HF were enrolled across six centres. After discharge, patients underwent 6 months of home monitoring using five household devices: a blood pressure monitor, body composition scale, activity monitor, fitness tracker, and smartphone. HF events were defined as HF rehospitalisation, unplanned outpatient visits, outpatient intravenous therapy for HF worsening, or cardiac death. To construct the alert system, a correlation-based anomaly detection approach was applied. For each patient, correlations among monitored parameters during the stable phase were used to define an individual baseline correlation structure. Deviations from this baseline structure were quantified as an anomaly score. Combinations of home monitoring parameters and clinical variables were compared to determine the optimal configuration for event prediction. The system generated three alert levels (green, yellow, and red), with red defined as a high-risk state indicating an HF event within 10 days. Results Analyzable data were obtained from 60 patients, among whom 8 HF events (13.3%) occurred. A model incorporating six parameters derived from the blood pressure monitor and body composition scale—night-time systolic blood pressure, night-time diastolic blood pressure, body weight, body fat percentage, skeletal muscle mass, and visceral fat level—demonstrated the highest predictive accuracy for HF events, with a sensitivity of 71.4%, specificity of 93.6%, and an area under the receiver operating characteristic curve of 0.80. In the validation dataset, red alerts preceded major HF events, including HF rehospitalisation and cardiac death, and were observed prior to clinical deterioration. Conclusion(s) A home telemonitoring-based alert system using only two simple, commercially available devices—a blood pressure monitor and a body composition scale—enabled early detection of HF events. This non-invasive and clinically implementable strategy represents a practical approach for improving HF management.