Sleep heart rate variability assists the automatic prediction of long-term cardiovascular outcomes

Highlights

• Two strategies for the segmentation of HRV signals were proposed and compared.

• The Surrounding strategy overwhelms the Truncating one in RF based REM classifier.

• A symmetric surrounding window with 390 RR intervals is optimal for REM detection.

Abstract

Objective

We aimed to investigate the association between sleep HRV and long-term cardiovascular disease (CVD) outcomes, and further explore whether HRV features can assist the automatic CVD prediction.

Methods

We retrospectively analyzed polysomnography (PSG) data obtained from 2111 participants in the Sleep Heart Health Study, who were followed up for a median of 11.8 years after PSG acquisition. During follow-up, 1252 participants suffered CVD events (CVD group) and 859 participants remained CVD-free (non-CVD group). HRV measures, derived from time-domain and frequency-domain, were calculated. Regression models were created to determine the independent predictor for long-term CVD outcomes, and to explore the association between HRV and CVD latency. Furthermore, based on HRV and other clinical features, a model was trained to automatically predict CVD outcomes using the eXtreme Gradient Boosting algorithm.

Results

Compared with the non-CVD group, decreased HRV during sleep was found in the CVD group. HRV, particularly its component of high frequency (HF), was demonstrated to be independent predictor of CVD outcomes. Moreover, normalized HF was positively correlated with CVD latency. The proposed prediction model achieved a total accuracy of 75.3%, in which sleep HRV features served as a supplement to the well-recognized CVD risk factors, such as aging, adiposity and sleep disorders.

Conclusions

Association between sleep HRV and long-term CVD outcomes was demonstrated here, suggesting that altered HRV during sleep might occur many years prior to the onset of CVD. Machine learning models, combining sleep HRV and other clinical characteristics, should be promising in the early prediction of CVD outcomes.

Introduction

Cardiovascular disease (CVD) is a major cause of mortality, claiming 33% of all deaths worldwide [1]. Early detection and control of CVD risk factors are therefore greatly encouraged for health management. Heart rate variability (HRV) is a useful term which is widely applied to describe the variation of intervals between two successive heart beats [2], such intervals are often called RR intervals. Since first was proposed, HRV has been rapidly adopted as a non-invasive method to study the cardiac autonomic modulation [3]. Evidences of an association between HRV and CVD, such as myocardial infarction [4,5], stroke [6], angina [7,8], coronary heart disease [9], coronary artery disease [[10], [11], [12]] and sudden cardiac death [13], have been reported. Furthermore, studies have put forward that HRV has predictive value for CVD outcomes [[14], [15], [16]]. For the general population, reduced HRV has a high correlation with incident coronary heart disease and death.

Among studies focused on the association between HRV and CVD, HRV signals used were usually acquired during daytime on those awake participants. Sleep, a totally different physiological condition from daytime awareness, constitutes a fundamental behavioral mechanism for all living organisms. For humans in particular, numerous evidences show that sleep is vital on maintaining physical health [17,18], cognitive function [19,20], recovery [21], memory [22], mood [23] and daytime functioning [24,25]. Furthermore, sleep, or sleep-related mechanisms, impose regulatory control over the cardiovascular system, since modulation of autonomic nervous system (ANS) are profoundly influenced by the sleep-wake cycle [26,27]. Eguchi et al., demonstrated that HRV during sleep was independently associated with an increased risk of CVD in patients with type 2 diabetes [28]. Vanoli et al., found that sleep HRV was highly relevant to the identification of autonomic derangements which may account for a higher risk of lethal events after myocardial infarction [29]. Recently, reduced parasympathetic modulation during sleep been reflected by the high frequency component of HRV was reported to be one potential mechanism underlying the increased prevalence of CVD among veterans with posttraumatic stress disorder [30]. Although the association between HRV and CVD has been well recognized, it is still largely unknown that whether HRV features, especially during sleep, can assist the prediction of the occurrence of CVD events after years of latency.

Several assessment systems for CVD risk have been proposed to predict individual CVD events, such as Framingham risk score [31], Reynolds risk score [32,33], QRISK2 risk score [34] and the prediction algorithm which is recommended by the American Heart Association/American College of Cardiology (ACC/AHA) [35]. Typical risk factors in these systems include age, systolic blood pressure, total and high-density lipoprotein cholesterol, smoking, hypertension and diabetes status. By means of such assessment systems, a large number of individuals at risk of CVD fail to be detected while some not at risk are given preventive treatment unnecessarily [36], new approaches are therefore still in demand to improve the accuracy of CVD prediction. Machine learning (ML) is a subset of artificial intelligence in the field of computer science that allows computers to use data to learn [37]. ML approaches have been widely applied in disease diagnosis and prognosis which achieving satisfied accuracy [38]. Recently, by adopting routine clinical data [36], ML algorithms were employed in a large-scale study to predict the first CVD event after 10 years. In comparison with the AHA/ACC risk prediction algorithm, results shows that the accuracy of CVD risk prediction is significantly improved by the application of ML [36]. However, to our best knowledge, few study employs HRV indices in ML models for CVD event prediction. Therefore, more explorations are required in predicting CVD outcomes automatically by using ML algorithms and HRV features.

It's worth noting that sleep is not a stable, but rather a complex process. For general population, a nocturnal sleep comprises four to six sleep cycles, which normally begin with light sleep, continue to deep sleep and end in rapid eye movement (REM) sleep [39]. According to the rules introduced by Rechtschaffen and Kales (R & K rules) [40], non-REM (NREM) sleep can be further classified into four stages (Stage 1, 2, 3, 4), and the current American Academy of Sleep Medicine rules combined Stages 3 and 4 and termed it N3 [41]. During sleep, ANS function is influenced by sleep state [26], resulting in an alteration of HRV across different sleep stages [[42], [43], [44]]. For healthy adults, HRV was observed to be decreased during NREM sleep with augmented parasympathetic modulation, and increased during REM sleep with a reduction in parasympathetic modulation [42]. Therefore, it is a rational way to comprehensively evaluate HRV in different sleep stages when using sleep HRV to predict CVD outcomes.

In the present study, we retrospectively analyze HRV data derived from an open-access database. On one hand, we target to investigate whether there is an association between sleep HRV and long-term CVD outcomes. On the other hand, we aim to find out whether ML model based on sleep HRV data and clinical characteristics can predict long-term CVD outcomes.