Causal Inference Based Hierarchical Regression Model for Cuffless Blood Pressure Estimation.

Area of Science:

• Biomedical Engineering
• Physiological Measurement
• Data Science in Healthcare

Background:

• Accurate cuffless blood pressure (BP) estimation is crucial for continuous monitoring.
• Existing methods often rely on correlations between physiological signals and BP, potentially overlooking confounding factors.
• Pulse transit time (PTT) derived from ECG and PPG signals is a common feature for BP estimation.

Purpose of the Study:

• To introduce causal inference for identifying confounders in PTT-based cuffless BP estimation.
• To propose a novel hierarchical regression model to mitigate the impact of identified confounders.
• To enhance the accuracy and reliability of noninvasive BP monitoring.

Main Methods:

• Utilized fast causal inference and causal generative neural networks to construct a causal graph.
• Identified key confounders influencing the relationship between PTT and BP.
• Developed and applied a hierarchical regression model incorporating the identified confounder (heart rate) for BP estimation.

Main Results:

• Heart rate was identified as the primary confounder affecting the PTT-BP relationship.
• The proposed hierarchical regression model achieved a mean absolute difference (MAD) of 6.29 mmHg for systolic BP and 5.03 mmHg for diastolic BP.
• The hierarchical model significantly outperformed traditional linear regression models.

Conclusions:

• Causal inference is a valuable tool for uncovering hidden relationships in physiological signal analysis.
• Accounting for confounders like heart rate through hierarchical modeling substantially improves cuffless BP estimation.
• This approach offers a promising pathway towards more accurate and robust noninvasive BP monitoring devices.