[Paper Review] Poincaré parameters and principal component analysis of Heart rate variability of subjects with health disorder

Sobhendu Kumar Ghatak, Subhra Aditya|arXiv (Cornell University)|Feb 28, 2018

Heart Rate Variability and Autonomic Control1 references7 citations

TL;DR

This study applies lagged Poincaré plot analysis, principal component analysis (PCA), and auto-correlation to short-term ECG data from diabetic, hypertensive, and healthy subjects. It reveals significantly reduced SD1, SD2, and SD12 parameters in diseased groups, with distinct curvature trends in SD12 versus lag—key findings validated by PCA, which clearly separates the three groups in a 2D principal component space, indicating enhanced discrimination of autonomic dysfunction beyond conventional HRV metrics.

ABSTRACT

Heart rate variability,important marker for modulation of autonomic nervous system is studied for diabetic,hypertensive and control group of subjects.Lagged Poincaré plot of heart rate(HR),method of principal component analysis and auto-correlation of HR fluctuation are used to analyze HR obtained from ECG signal recorded over short time duration.The parameters $(SD1)$,$(SD2)$ and their ratio $(SD12)$,characterizing the Poincaré plot reveal a significant reduction of their values for diabetic and hypertensive subjects compared to the corresponding results of control one.The slope and the curvature of the plot of these parameters with lagged number exhibit similar trend.In particular,the curvature of $(SD12)$ for the control group differs widely from that of other groups.The principal component analysis is used for analysing multi-dimensional data set resulting from the Poincaré plot for all subjects.The correlation matrix points out significant correlation between slope and curvature.The analysis demonstrates that three groups are well separated in the domain of two significant principal components.The auto-correlation of HR fluctuation exhibits highly correlated pattern for subjects with health disorder compared to that of control subject.

Motivation & Objective

• To investigate whether lagged Poincaré parameters and PCA can differentiate heart rate variability (HRV) patterns in diabetic, hypertensive, and healthy subjects.
• To assess the utility of Poincaré plot parameters (SD1, SD2, SD12) and their growth with lag in detecting autonomic nervous system (ANS) imbalance.
• To evaluate whether principal component analysis of multi-dimensional Poincaré-derived variables enhances separation between health-disorder groups.
• To examine auto-correlation of RR interval fluctuations as a marker of non-random, pathological dynamics in HRV.

Proposed method

• Lagged Poincaré plots were generated by plotting RRi against RRi+m for increasing lag m (up to 8), capturing temporal correlations in beat-to-beat heart rate dynamics.
• Key parameters SD1 (short-term variability), SD2 (long-term variability), and their ratio SD12 were extracted from the lagged Poincaré plots to quantify HRV distribution.
• Principal component analysis (PCA) was applied to the multi-dimensional dataset of Poincaré parameters to identify dominant patterns and reduce dimensionality while preserving variance.
• The correlation matrix of Poincaré parameters revealed significant interdependencies, particularly between slope and curvature of SD12 with lag.
• Auto-correlation of RR interval fluctuations was computed as a function of lag m to assess persistence and oscillatory behavior in HR dynamics.
• Data were shuffled five times to test for randomness; the resulting correlation patterns were compared to original data to validate non-random structure.

Experimental results

Research questions

• RQ1 Do lagged Poincaré plot parameters (SD1, SD2, SD12) differ significantly between diabetic, hypertensive, and control subjects?
• RQ2 How do the slope and curvature of Poincaré parameters (especially SD12) with increasing lag m distinguish the three subject groups?
• RQ3 Can principal component analysis effectively separate the three groups based on multi-dimensional Poincaré parameters?
• RQ4 Is there a measurable difference in the auto-correlation of RR interval fluctuations between healthy and diseased subjects?
• RQ5 To what extent do the derived parameters (SD1, SD2, SD12) reflect autonomic nervous system imbalance in cardiovascular disorders?

Key findings

• Diabetic and hypertensive subjects showed significantly reduced SD1, SD2, and SD12 values compared to the control group, indicating impaired short- and long-term HRV.
• The curvature of SD12 with respect to lag m was markedly different in the control group compared to both diabetic and hypertensive groups, suggesting a unique dynamic signature in healthy autonomic regulation.
• The slope and curvature of SD12 exhibited consistent trends across all groups, with the control group showing the most distinct curvature profile.
• PCA revealed that the first two principal components effectively separated the three groups in a 2D space, demonstrating strong discriminative power of the Poincaré-derived parameters.
• The first four principal components, associated with significant eigenvalues, preserved most of the data variability, with the dominant component heavily weighted on SD12.
• Auto-correlation of RR fluctuations showed a finite, decaying, and often oscillatory pattern in diseased subjects, contrasting with near-zero, random-like correlation in shuffled data, indicating non-random, pathological dynamics in HRV.