Intradialytic blood pressure pattern recognition based on density peak clustering.

Area of Science:

• Nephrology
• Data Science
• Biostatistics

Background:

• End-stage renal disease (ESRD) necessitates hemodialysis (HD), where intradialytic blood pressure (IBP) monitoring is crucial for patient safety and prognosis.
• Existing studies on IBP patterns are limited and rely on subjective classifications, potentially biasing results.

Purpose of the Study:

• To develop an objective approach for identifying and classifying intradialytic blood pressure patterns in ESRD patients undergoing hemodialysis.
• To assess the clinical and prognostic significance of these identified IBP patterns, particularly concerning cerebrovascular event risk.

Main Methods:

• Employed dynamic time warping (DTW) to quantify similarity between IBP data series.
• Utilized density peak clustering algorithm (DPCA) to identify distinct IBP patterns.
• Constructed random survival forest (RSF) models to evaluate the impact of IBP patterns on prognosis.

Main Results:

• Identified five distinct blood pressure patterns using DPCA.
• Inclusion of IBP patterns in RSF models improved predictive accuracy by 3.7-6.3%.
• BP patterns demonstrated critical clinical and prognostic significance for cerebrovascular event risk.

Conclusions:

• The proposed objective clustering approach effectively identifies clinically relevant IBP patterns in ESRD patients.
• IBP patterns are significant predictors of prognosis, especially for cerebrovascular events.
• This methodology is generalizable to other time-series data within electronic health records (EHRs).