Present trends and future directions in the analysis of cardiovascular variability

The analysis of heart rate variability (HRV) was born as a separate research area more than three decades ago, following the path opened by the first pioneering works that described the spontaneous HR dynamics in terms of spectral powers [1,2]. The possibility that such a simple technique could provide information on cardiovascular regulatory mechanisms raised high hopes. In the following years this produced thousand of scientific articles investigating physiological and pathological aspects of cardiovascular homeostasis associated with HRV spectral patterns [3]. However, despite the important contributions to the physiological understanding of cardiac control mechanisms, methods of HRV spectral analysis have so far failed to find clear clinical applications. Although a few articles showed a prognostic value for HRV, either after myocardial infarction or in heart failure patients [4,5], the prognostic value of HRV was not confirmed by other articles. Furthermore, a number of studies emphasized that multiparametric approaches, for example linking blood pressure variability (BPV) and HRV to estimate baroreflex sensitivity, might give better results [6,7]. In fact, there has been a growing awareness that in order to be used as an effective clinical tool, information from HRV spectral powers should be properly integrated, so as to provide a more complete picture of the overall cardiovascular regulation and of the interactions among physiological systems. This requires rethinking traditional methods of data collection and analysis. A first possibility is to redesign the data collection phase by including other variability signals, thus completing the information provided by HRV. For instance, as mentioned above, the recent availability of devices providing beat-to-beat arterial BP measures noninvasively and continuously for long periods allows the easy combination of HRV analysis with BPV analysis and the quantification of their in different ways [8,9]. Potential applications include the clinical evaluation of congestive heart failure and diabetic patients [10,11]. Moreover, the simultaneous recording of HR data with arterial pressure and respiratory signals during overnight polysomnographies may also allow the clinical evaluation of the effects of sleep-related breathing disorders and of their treatment from HRV and BPV analysis [12–14]. A nice example of the integration of HRV analysis with the analysis of variability in other physiological signals is also given in the study by Elstad et al.[15] appearing in the present issue of this journal. The study focuses on the identification of mechanisms relating the Mayer waves with the low-frequency power of HR. Mayer waves are BP fluctuations occurring at around 0.1 Hz in humans [16]. These BP fluctuations are associated with similar fluctuations in HR, as indicated by the high coherence between BPV and HRV in the low-frequency band (i.e. the frequency band centered around 0.1 Hz). Although these fluctuations have been deeply studied, their origin is still controversial. In fact, experimental data in humans cannot be unequivocally interpreted, suggesting both a central origin [17] as well as a peripheral origin [18,19], probably caused by a resonance in the baroreflex loop. Moreover, independent of their origin (central or peripheral), it is still unclear whether HR oscillations in the low-frequency band contribute to the phenomenon of Mayer waves by generating or by buffering BP oscillations. Understanding these mechanisms may help to better interpret the clinical relevance of these components of cardiovascular dynamics and to more efficiently use them for properly describing the patients' conditions in clinical settings. Particularly, the study presented by Elstad et al.[15] in this issue of the journal is aimed at investigating the influence of low-frequency fluctuations of HRV on the Mayer waves of BP. Do low-frequency powers of HR contribute to the Mayer waves (feed-forward mechanism) or rather do they buffer BP waves (feedback mechanism)? To answer this question, the authors performed a series of technically demanding recordings in healthy volunteers, collecting not only beat-to-beat values of HR and BP but also beat-to-beat values of stroke volume by means of an ultrasound Doppler device, from which beat-to-beat values of cardiac output and total peripheral resistances were derived. This approach provided the authors with a rich picture of the dynamics of the main cardiovascular variables possibly involved in the genesis of the Mayer waves. Moreover, experiments were performed during supine rest and mild orthostatic activation, with and without selective autonomic blockade. In this way the authors could somehow separate the influences of those variables more directly affecting the Mayer waves. Finally, quantification of phase angles between HR and cardiac output, between cardiac output and BP and between BP and total peripheral resistance allows an evaluation of the feedback or feed-forward nature of HRV–BPV interactions. The results of this article [15] reveal complex relationships among the cardiovascular variables considered. Particularly, these data suggest that changes in total peripheral resistance directly contribute to the Mayer waves, because they are always in phase with BP oscillations. During mild orthostatic stress, cardiac output is in counterphase with BP. As the HR is in phase with cardiac output, this would suggest a potentially buffering role of HR fluctuations in the low-frequency band. However, removal of HRV by vagal blockade did not affect fluctuations of BP. Vagal blockade also did not affect total peripheral resistance. Therefore, the authors conclude that during a mild orthostatic activation the low-frequency powers of HRV are not able to buffer efficiently the Mayer waves induced by oscillations in total peripheral resistances. The study by Elstad et al.[15] clearly emphasizes the importance of including more cardiovascular variables in HRV studies. This would allow a better understanding of the cause–effect relationships among variables in physiological and clinical settings. However, a limitation of most studies, that by Elstad et al.[15] included, is that, when considering the interaction between fluctuations in different parameters, focus on the phase angle between their respective fluctuations does not univocally indicate the time delay between variables. Particularly, no cause–effect relationship can be inferred from the sign of the phase difference. Research in this field is attempting to overcome this problem in different ways. One possibility is to fit mathematical models with the measured data from which one can derive information on the relationship between variables. For instance, the use of bivariate auto regressive moving average models allows separating feedback and feed-forward components of the relation between HRV and BPV by quantifying their mutual interaction [20,21]. A different approach consists in the evaluation of causality, a way to quantify the dominant temporal direction among cardiovascular variables [22–24]. Causality, in technical terms, is strictly related to time irreversibility, which means that the dynamics of cardiovascular variables, like HR, may look different if the temporal axis is reversed [25,26]. Therefore, irreversibility is in turn associated with asymmetrical dynamics of the beat-to-beat series [27,28]. Asymmetry in HR dynamics is also related to the fact that the capacity of the autonomic nervous system to accelerate the heart rhythm differs from the capacity to decelerate it [29]. These topics, currently investigated by different groups, appear promising because they allow including aspects of HRV dynamics which cannot be addressed properly by traditional spectral methods. Another aspect of HR dynamics that cannot be properly described by traditional spectral methods is the degree of ‘irregularity’, or ‘unpredictability’, of the HR series. To describe this aspect, a method for estimating entropy of physiological signals, called approximate entropy (ApEn) [30] was proposed in the 1990s. ApEn is based on the calculation of the probability that patterns of HR data repeat themselves. If the probability to repeat patterns is low, the series appears ‘irregular’ and it will be difficult to predict its behaviour. The lower the probability of finding repetitive patterns, the higher will be the entropy of the series. Research in this field led to variants of entropy estimators. For instance, sample entropy improved ApEn by excluding self-matches in the calculation of repetitive patterns [31]. Correction for self-matches was also introduced by Porta et al. by proposing the corrected conditional entropy estimator [32]. Until now, the more promising clinical applications of entropy-based HRV analysis have regarded the prediction of episodes of atrial fibrillation [33–35]. More recently, multiscale entropy has extended the concept of unpredictability by quantifying the HR irregularity separately over the different temporal scales that constitute the HR dynamics [36]. Multiscale entropy was able to point out significant differences between heart failure patients and healthy controls in the HR irregularity [36], a result which was not possible to obtain with ApEn [37,38]. Recent advancements in HRV analysis have also affected the assessment of HR self-similarity, another aspect not easily quantifiable by spectral methods. Self-similar series behave like geometric fractals. They can be split into small fragments and still each fragment may show fluctuations somehow similar to the original series, if properly rescaled. This behaviour characterizes signals, like HR, generated by ‘complex’ systems. In fact, anatomical fractal structures in the heart and in the vasculature, and integration of local and global regulatory mechanisms, introduce ‘fractal’ components in HR dynamics which are not characterized by a specific temporal scale. This means that the HR series may look somehow similar when plotted at different time scales, if the vertical axes are rescaled by a certain ‘stretching factor’. Detrended fluctuations analysis (DFA) [39] is usually applied to derive one-scale or, more often, two-scale coefficients (a short-term and a long-term coefficient). These coefficients are associated with the HR fractal dimension and define the ‘stretching factors’ of the HR self-similarity. Clinical studies investigated the prognostic power of the DFA short-term coefficient in postinfarction patients [40] and in heart failure patients [41] (see a detailed review in [42]). However, it has been recently suggested that a single DFA coefficient cannot reliably reflect the autonomic influences on HR [43], and also that the ‘two-coefficient’ model may be an oversimplification of a more complex phenomenon, prone to arbitrary selections of the temporal scales over which the DFA coefficients are defined. This has led to propose approaches aimed at quantifying the HR self-similarity by means of a full spectrum of DFA coefficients [44,45]. With this approach, the autonomic influences on the HR fractal structure are clearly and consistently defined [46]. Moreover, it has been shown that a whole spectrum of DFA coefficients classify congestive heart failure patients significantly better than possible by employing the more traditional short-term and long-term coefficients [47]. The spectrum of DFA coefficients describes the HR self-similarity as composed by fractal regimes separable over different temporal scales. When the HR is assumed to be composed of more complex mixtures of ‘interwoven’ fractal processes, different approaches based on the calculation of multifractality are preferred. It has been recently shown that the HRV dynamics of normal and congestive heart failure patients (but not of patients in atrial fibrillation) are compatible with the hypothesis of multifractality and are characterized by different multifractal structures [48]. The actual trends in the analysis of HR dynamics and their more interesting applications in clinical settings are summarized in Table 1[21,24,25,28,29,33–36,40,41,47,48].Table 1: Actual trends in methods of heart rate variability analysis and more recent applications on clinical dataIn conclusion, the hopes raised originally by the first studies on HRV that HR spectral analysis could provide effective clinical tools for prognosis and risk assessment of cardiac patients are still alive. These hopes, however, are more likely to be fully accomplished if information from HR dynamics can be integrated by other signals, so as to have a more complete picture of the whole spontaneous cardiovascular variability, and if the traditional spectral methods are combined with more sophisticated mathematical tools for assessing the complex interactions among the regulatory mechanisms underlying cardiovascular homeostasis.