Machine learning-driven integration of 24-hour ambulatory blood pressure and its variability

Area of Science:

• Cardiology
• Biostatistics
• Data Science

Background:

• Twenty-four hour ambulatory blood pressure monitoring (24-hour ABPM) is crucial for hypertension assessment.
• Interpreting complex 24-hour ABPM data and its variability poses challenges.
• Integrative analysis is needed to link BP patterns with cardiovascular outcomes.

Purpose of the Study:

• To investigate if a clustering algorithm can integrate 24-hour ABPM data and variability.
• To identify distinct patient clusters associated with cardiovascular (CV) outcomes.
• To validate the clinical utility of clustering for hypertension management.

Main Methods:

• Utilized dynamic time warping and k-medoids clustering on 24-hour ABPM (HR, SBP, DBP) data.
• Included 1344 community-dwelling individuals with clinical data and adverse outcomes.
• Validated clusters using an external cohort (n=1219) and clinical characteristics.

Main Results:

• Identified 4 distinct clusters based on BP patterns and variability.
• Cluster 1: Favorable CV risk profile with younger individuals and lower BP.
• Cluster 4: Poorer CV risk profile with older individuals, higher BP, and increased BP variability; significantly higher CV event risk (HR: 1.63) compared to Cluster 1.

Conclusions:

• Time series clustering of 24-hour ABPM effectively integrates BP and variability.
• This approach facilitates the interpretation of complex BP data for risk stratification.
• Clustering offers a novel method for identifying high-risk hypertension patients and improving CV outcomes.