Data-Driven Characterization of Heart Rate Dynamics During Sleep Using Functional Data Analysis

Abstract

Introduction

Obstructive Sleep Apnea (OSA) is a highly prevalent sleep disorder characterized by recurrent upper airway obstruction during sleep, leading to intermittent hypoxia and autonomic nervous system dysregulation. Heart Rate Variability (HRV), derived from RR interval series in ECG recordings, has been widely studied as a non-invasive marker of autonomic dysfunction in OSA. However, conventional HRV analyses rely on summary statistics that may fail to capture the full temporal structure of the signal.

Objective

This pilot study explores the application of Functional Principal Component Analysis (FPCA) to RR interval series extracted from polysomnographic ECG recordings, with the aim of identifying functional features associated with OSA severity and respiratory event characterization.

Methods

We retrospectively analyzed ECGs from full diagnostic polysomnographies from patients evaluated at the IRCCS Istituto Auxologico Italiano. Raw overnight ECGs were denoised and then segmented into non-overlapping event-locked epochs centered on individual respiratory events. Additionally, non-event ECG segments were extracted in order to utilize them as a control group. For each segment, RR interval series were extracted using a validated delineation algorithm for R peaks detection. Each RR series was then modeled as a continuous function over time using Functional Data Analysis (FDA) techniques and Functional Principal Component Analysis (FPCA) was applied in each setting to extract the leading modes of variation in RR series. The resulting functional principal components (FPCs) were examined in relation to AHI values and respiratory event typology.

Results

The event-locked analysis (n = 281) was conducted under two perspectives: discrimination among respiratory event typologies and comparison between event and non-event segments. Scatter plots of the four leading FPCs (which accounted for 48.8%, 12.5%, 9.5% and 7.3% of the total variance, respectively) revealed no clearly discriminative functional structure between event typologies. This is arguably interpretable, as both event types may elicit highly similar cardiac responses, producing largely overlapping functional signatures. Notably, the first functional principal component predominantly described fluctuations around the mean RR value, effectively capturing the overall sympathovagal balance, rather than event-specific modulation patterns. A similarly inconclusive picture emerged from the event vs non-event comparison, which is in this case more striking. Despite the autonomic perturbation associated with respiratory events being expected to produce a detectable functional signature relative to baseline, the FPC scatter plots showed substantial overlap between the two groups. High intra- and inter-individual variability in RR dynamics appears to dominate the functional representation, masking any systematic difference attributable to the presence of a respiratory event.

Conclusion

These preliminary findings suggest that naive application of FPCA to raw RR series may be insufficient to extract OSA-related functional structure, due to the dominance of physiological variability unrelated to respiratory events. Rather than representing a negative outcome, these results provide methodological insight into the limitations of direct functional decomposition and motivate the exploration of individual-centered approaches to better isolate the autonomic signature of OSA from background variability. Further analyses are ongoing.