Nocturnal Blood Pressure Estimation from Sleep Plethysmography Using Machine Learning.

Area of Science:

• Biomedical Engineering
• Cardiovascular Research
• Machine Learning in Healthcare

Background:

• Elevated nocturnal blood pressure (BP) is a significant risk factor for cardiovascular disease (CVD) and mortality.
• Traditional cuff-based BP monitoring is impractical for sleep assessment.
• Cuffless BP monitoring using machine learning presents a promising alternative.

Purpose of the Study:

• To develop and evaluate a machine learning algorithm for predicting nocturnal BP.
• To utilize single-channel fingertip photoplethysmography (PPG) for cuffless BP monitoring during sleep.
• To assess the feasibility of this method in healthy adults.

Main Methods:

• Sixty-eight healthy adults underwent overnight polysomnography (PSG), fingertip PPG, and ambulatory blood pressure monitoring (ABPM).
• Pulse morphology features were extracted from PPG waveforms.
• Random forest models were employed to predict night-time systolic (SBP) and diastolic (DBP) blood pressure.

Main Results:

• The optimal prediction window length was determined to be 7 seconds.
• The model achieved a mean absolute error (MAE) of 5.72 mmHg for SBP and 4.52 mmHg for DBP.
• High correlation coefficients (0.87 for SBP, 0.86 for DBP) were observed between predicted and measured BP.
• The stiffness index was identified as the most crucial PPG feature for BP prediction.

Conclusions:

• Machine learning-based nocturnal BP prediction using fingertip PPG is feasible in healthy adults.
• The developed cuffless method accurately captures the complex relationship between PPG and BP during sleep.
• This technology offers a scalable, convenient, economical, and non-invasive approach for continuous blood pressure monitoring.