Noninvasive Multiparameter Monitoring for the Detection of Decompensated Heart Failure: Exploratory Study.

Area of Science:

• Biomedical Engineering
• Cardiology
• Digital Health

Background:

• Current remote patient monitoring for heart failure (HF) relies on manual data interpretation.
• Automated multiparameter predictive models (MPMs) can enhance early detection of decompensated HF and reduce healthcare professional workload.
• Noninvasive, user-friendly devices are crucial for cost-effective, large-scale implementation of MPMs.

Purpose of the Study:

• To evaluate an MPM utilizing a wrist-worn device for detecting decompensated HF and mortality in unstable HF patients.
• To assess the contribution of photoplethysmography and accelerometer data to HF event prediction.

Main Methods:

• Seventeen unstable HF patients wore a wrist-worn device measuring heart rate (HR), interbeat intervals (IBIs), respiration rate (RR), activity counts (AC), and energy expenditure (EE) for 3 months post-discharge.
• Seven classifiers were evaluated, with the best model tested using leave-one-subject-out cross-validation.
• The combined endpoint included hospital readmission, outpatient diuretic dose increase, or HF-related death.

Main Results:

• Device-wearing compliance was 78%. Activity-related parameters (EE, AC) yielded the highest data quality (72-79%).
• High-quality data for HR (46%), IBI (29%), and RR (14%) were limited; sleep data quality was 1%.
• The optimal MPM achieved 97.2% specificity and 5.3% sensitivity for predicting HF deterioration within 2 weeks of an event (AUC=0.59).

Conclusions:

• The noninvasive wrist-worn MPM demonstrated high specificity but low sensitivity in predicting decompensated HF and mortality.
• Low sensitivity was attributed to extreme class imbalance and poor data quality (HR, RR, sleep) in this older HF cohort.
• Future studies require improved data fidelity and larger cohorts to enhance predictive performance for HF events.