The human body consists of many different interacting systems. The brain would not survive without circulation, and circulation is only possible when energy metabolism is working. Eating food is only safe when the liver extracts the toxic substances and the immune system defends against pathological germs, and so on. In reductionist bioscience, the human physiology is investigated by identifying its parts and focusing on one of it. In controlled experiments, one factor is changed and every other possible controlled as much as possible. This approach has been extraordinarily successful in understanding basic principles of human (or animal) physiology. In principle, the reductionist approach assumes that if scientists investigate every part of the human body and its possible interactions with other parts and when they put all together they will understand human physiology.

This approach does not necessarily work out. Even reductionists agree that a brain will die within 3 min when circulation breaks down. However, the reductionist approach does not appreciate the tight interconnection of different parts of the body to function in real time. The basic physiological approach based on research in the nineteenth century is still valid when looking at some general functions. Heart function can be described by a linear equation where cardiac output is multiplied with heart frequency. The view changes, however, when the level of detail is increased. Every heart beat is at least slightly different; cardiac output is influenced by factors as simple as preload, afterload, or systemic resistance, or as complicated as a plethora of humoral factors, a variety of efferent signals and the condition of the heart muscle. Neural inputs are never the same; their patterns are depending on afferent signals from the heart to different brain centers and again humoral signals. These are modified by energy metabolism, immunological signal cascades, the internal state of the brain itself. It is possible to identify hundreds of different factors which probably influence the cardiac output in various ways. However, it is also possible to turn it around. The cardiac cycle has effects on brain function beyond simply providing enough blood flow. Oscillations in the brain are coupled to the heartbeat and oscillations in other systems. The state of the immune system has effects on the brain, which again has effects on the circulation system, apparent in sepsis, but relevant also in less dramatic situations. It turns out that the human body consists of closely interconnected parts which communicate closely in real time on the level of changes in time orders between milliseconds and hours, even days.

Another part of the picture regards the influence of perturbations. In a linear, reductionist world, a change in one condition will have a straightforward effect on the connected system. Increase adrenaline twice, and the heartbeat increases twice. Take away some blood, and the blood pressure decreases, take away twice as much, and the blood pressure decreases twice as much. The reality is different. A change of internal or external factor will often only induce minor changes. Losing a half liter of blood will probably induce small changes in blood pressure and slightly increase heart frequency. Removing more will eventually lead to sudden changes, at the end to a breakdown of circulation.

Meet another paradigm. The human body can be regarded as a system. A system can be defined as a set of interacting or independent components forming a whole. Each system has boundaries defining an inside and an outside. Systems again can consist of a set of systems (subsystems) interacting in the same way. This definition can be applied to physical entities, human bodies, or social constructions in the same way. Systems can be studied by breaking them up into parts as reductionists do, but the notion of the system includes that a system “is more than its parts” (1). Systems, in general, have some properties in common. They can be remarkably stable against perturbations, but they also can be remarkable fragile in case of a specific perturbation, leading to a sudden change. This possible change is frequently termed phase change, and the ability to change is termed emergence. Apparently, sudden system changes due to minor changes in perturbations are not linear, but non-linear. This means simply that a small perturbation might cause a major system change, or a significant perturbation only a minor modification. The advantage using this approach to the human body is that both a sophisticated understanding and sophisticated tools to investigate systems have been elaborated (2, 3).

Biomedical research in the last decades has reached an enormous level of detail. Independent of the area of investigation, the knowledge of genetics, translational mechanisms, intracellular pathways, receptor systems and their ligands, or immunological mechanisms have reached a level where it is challenging to understand the system as a whole. It turned out that simple systems are not as simple, physiological systems are best assumed to be more complex than at first apparent until it can be demonstrated otherwise (2). All physiological systems can be understood in a cybernetic way. They consist of control cycles using either positive or negative feedback mechanisms which were identified in the second half of the last century. Systems with more than three positive and negative feedback circles in combinations are often not predictable; they show complex behavior.

At the end of the last century, the increasing level of details provided by biomedical research provoked debates how to understand the whole system. Similar debates also occurred in other scientific disciplines. System theory and cybernetics gradually evolved to something which was termed complexity science although an exact definition was challenging and no unified theory existed. At the same time period, system theory evolved to dynamic system theory beyond other influences also with the help of inputs from chaos theory which in reality investigates the behavior of deterministic non-linear systems (3). Another evolving direction of investigations was based on increasing computational power which it made possible to simulate large artificial systems in real time. Agent base modeling, cellular automata, and genetic algorithms were developed. All these approaches made contributions to understand complex systems; main tools repeatedly used were an analysis of time series of signals and computer simulations (4).

In the beginnings of 2000s systems biology was established. It is a biology-based inter-disciplinary field of study focusing on complex interactions within biological systems on several levels—intracellular, intercellular, hormonal, macroscopic—using a comprehensive approach and as tools data mining, mathematical models, and time series analysis (4, 5). A time series is a sequence of observations made over time, in the case of heart-rate variability (HRV) heartbeats, but also other parameters (like blood sample results, EEG signals, etc.) (6). Time series analysis can help to identify hidden patterns and even causalities in physiological systems (7). Systems biology analyses different parts of the body as a kind of network (8). Most investigations are conducted on the subcellular level, looking on metabolic and genetic regulation networks and then analyzing them with the help of sophisticated network methods, which again are validated with experimental data. Network analysis or models can even be applicated in the case of unknown quantitative relationships or lacking precise data (e.g., By using Boolean networks) (9, 10).

This review is organized as follows. After a short introduction of HRV, some physiological systems with influence on HRV will be reviewed. In the next part, different paradigms of HRV are presented—HRV as time series analysis, as a proxy for the autonomic vegetative systems and as a proxy for a complex regulatory system. Finally, implications will be discussed.

The principles of HRV are simple. The heartbeat is measured, usually with the help of an ECG signal which is obtained with an adequate device. A minimum sampling rate between at least 250 and 500 Hz is recommended (11). QRS-distances are measured (called NN-distances) after identification of ventricular and supraventricular extrasystoles which usually are interpolated. Subjects with a high rate of extrasystoles or atrial fibrillation (AF) are usually not feasible for analysis [but see Ref. (12)]. At the end of the measurement period, a time series of milliseconds can be processed in time domain, frequency domain, geometrical measures, and different non-linear measures like fractal parameters or different calculations of entropy (13). The most common used parameters are the SD of NN Intervals (SDNN) and root means successive square difference (rMSSD), calculated by squaring of each NN interval, thus calculating the mean value and drawing the square root (14). Geometric methods are derived from sequences of NN intervals. Different geometrical methods include the 24-h histogram, the HRV triangular index, the triangular interpolation of NN interval histograms, and methods like the Poincaré-plot. Frequency domain (power spectral density) describes the periodic oscillations of the heart-rate signal, transformed into different frequencies areas, and returns numerical values about their relative intensity (14). Most frequently, the frequency-domain parameters are calculated non-parametrically with the fast Fourier transformation. The most used parameters are Total Power, VLF (very low frequency, <0.003–0.04 Hz), LF (low-frequency power, 0.04–0.15 Hz), HF (high-frequency power, 0.15–0.4 Hz), and the ratio LF/HF. HF is often understood as a proxy for the parasympathetic nervous system (PNS). It can be influenced by the frequency of breathing or pathological forms of respiration, but is reliable with normal respiration (15) and is to a certain grade similar to respiratory sinus arrhythmia (RSA) (16). Both the sympathetic nervous system (SNS) and the PNS modulate LF. High LF reflects often increased sympathetic activity. The LF/HF ratio might reflect the global sympathetic/vagal balance. VLF is probably influenced by the renin–angiotensin system and is also associated with the sympathetic activity (15, 17). Non-linear methods often focus on (self)similarities in the time series. Classical algorithms for the analysis of self-similarities are fractal methods. Different forms of entropy measures have been used (18). A standard for measurement and interpretation was published in 1996, and most studies afterward are based on it (11).

Time series show the variation of a parameter during a period. In complexity theory, they are considered as surrogates of the particular system, giving information about the state of the system. Typical time series can consist of a series of blood samples (e.g., CRP, creatinine, white blood cell counts), physical measurements like heartbeats, blood pressure, EEG; or movements like gait, eye blinking or variability of breathing patterns. Even short time series can also give information about a system. Usually, most properties of time series are ignored. They are summarized by simple statistical calculations like the mean and the SD. In the case of HRV, several linear and non-linear algorithms were eventually introduced. The research literature can be roughly divided into three areas. In the first, HRV parameters are used as a prognostic tool. Short-term or long-term results in trials are associated with (usually decreased) HRV parameters. This approach has been very successful providing studies where an analysis of only 2 min of heartbeat predicted all-cause mortality many years later, e.g., in Ref. (103–105). The second approach uses HRV as a proxy for the state of the vegetative system, based on the century-old notion of a balanced situation of the sympathetic and the parasympathetic part. The third approach relies on the complexity paradigm. It considers variability as a proxy for the health of a system and decreased variability as a sign of a pathological condition. In the next section, I will discuss the scientific evidence for associations between HRV and the state of the vegetative system, followed by a section where I present the third approach.

Heart-rate variability is very often used to describe the activity of the SNS and PNS. This is based on the assumptions that there is something like a general activity level of the ANS, that HRV parameters mirror this activity and that SNS and PNS are in the balance, meaning that high SNS activity is associated with low PNS activity and vice versa. If this is the case, it should be possible to measure consistent patterns in the activity of different vegetative efferents. But this supposed consistency has frequently not been found. Different other parts of the SNS have not necessarily the same kind of oscillations as the cardiac part of SNS if they have oscillations at all. Cutaneous vasoconstrictor fibers, sudomotor fibers, adrenaline-regulating adrenal preganglionic neurons, and nerves supplying the brown adipose have not oscillations as observed in other parts of the SNS (106–109).

The sympathetic output might also be as similar in different parallel measured subsystems (110). The physiologist William Cannon propose at the beginning of the last century a model of autonomic control, where either sympathetic activation is high and parasympathetic activation is low or vice versa (111). In contradiction to this frequently used model evidence has been provided that descending influences from higher neural systems can trigger all patterns of changes in the SNS and PNS, whether reciprocal, independent or even co-active (112, 113). Already early observations indicated that both divisions of the vegetative nerve system could be similar active and work together with the somatic motor system to regulate most behavior, both in normal and in emergency conditions (114). Even when either the SNS or PNS are predominant in the control of an organ or anatomical region (110) and even when the sympathetic and the parasympathetic division have opposed effects on these areas, the balance is more complicated. Coactivation or a degree of opposing activations patterns can both be important to keep homeostasis when external conditions change (22). A major challenge is to interpret experimental approaches which often use isolated organ systems in rather artificial experiments, where other systems are lacking.

The classical description of a stress response of the SNS includes increased heart beat frequency, vasoconstriction, piloerection, and pupillary dilatation (115). However, stress response can be independent of SNS activation.

The increased muscle blood flow, for instance, is triggered by activation of a cholinergic vasodilatation pathway, with only minor involvement of adrenergic pathways (115, 116). Also during the diving reflex, simultaneous activation of the sympathetic and parasympathetic part of the ANS has been observed. When the head is under water, it causes a massive activation of trigeminal afferents nerve fibers which again initiates an intense and simultaneous reaction of SNS and PNS. The SNS triggers massive vascular constriction everywhere (except brain and heart), the vagal activation triggers severe bradycardia and decreases cardiac contractility. The meaning of the diving reflex is to reduce energy demand in a stress situation with an anticipated lack of oxygen in the body (117, 118). Interestingly, the intense simultaneous activation of SNS and PNS is associated with a dramatic loss of fractal properties of the HRV signal (119). Another example is the simultaneous pathological influence of SNS and PNS on AF (120), probably underestimated because HRV usually cannot be used as a tool in these patients. A daily observation in the operation theater is the change of heart rate caused by a sudden nociceptive input in a patient with superficial anesthesia: skilled anesthetists notice a sudden moderate decline of heart rate following by a marked increase some seconds later, most probably caused by simultaneous activity increase of the SNS and PNS, most marked observed in strabism surgery (121). Hundred years after Walter Cannon’s insights, we should accept that high SNS activity does not necessarily imply a low PNS activity.

Another important question is if the different parameters of HRV mirror the state of the ANS. It has been shown, that representations of ANS in the brain and brain centers involved in HRV are mainly the same (122–125). However, does HF correlate with the parasympathetic tone? The presented evidence often consists of pharmacologic studies where the sympathetic system is blocked with, e.g., atropine (126–128). In a complex experiment, the relationship between HRV parasympathetic activity was described by a function with an ascending part that goes over to a plateau level (129). These results have been challenged, and some authors argue that HF is a possible, but not optimal index (130), or that it only works out when respiration is controlled (131). Similar in the case of LF and sympathetic activity: several studies confirm the relationship (128, 132, 133), but not in all subjects (134) and some results were challenged (135). Contradictory results are, e.g., when an intense training session in healthy persons (which should usually increase sympathetic activity) is associated with a decrease of HRV parameters (136). High dose atropine is expected to block completely vagal parasympathetic activity and according to Cannon’s model increase sympathetic activity, but it nearly all LF power (and HF power) disappeared in an investigation (126, 127). Most of the studies regarding HRV components and different parts of ANS were conducted more than 20 years ago. Since then, the debate has calmed down, and these associations have been assumed by most authors using HRV. The relation between HRV parameters and ANS does certainly exist, but the reality is probably more complicated and more research is needed.

Kauffman, affiliated with the renowned Santa Fe Institute for Complexity Research, proposed the notion of life being generally at the edge of chaos (137, 138). Based on theories of complex (adaptive) systems, life is defined being at the frontier between mathematical chaotic behavior and (some kind of) order. Order is here defined as the possibility to estimate the behavior of the system based on its internal states during the last period. The same argumentation is used of some authors arguing theoretically that “health” is defined not being in complete thermodynamic equilibrium; too close proximity (decreased variation, too little energy dissipation, low entropy) or too far distance (increased variation and energy dissipation, high entropy) would indicate pathology (139). Decreased HRV parameters are usually considered as pathological, according to a notion of Goldberger (140). Changed HRV variability, as seen in elderly individuals is explained as reduced number of system components and reduced coupling between elements in a complexity theories paradigm (141, 142). Many studies seem to support this paradigm, but there are also some contradictory reports where increased variability is associated with illness. In endocrinological diseases, HRV can be increased compared to healthy controls. In patients with acromegaly (caused by increased levels of growth hormone due to hypophysis adenoma), the amount of growth hormone release over 24 h correlates with higher approximate complexity (143). Also in patients with Cushing syndrome, approximate entropy is greater when ACTH and cortisol concentration levels are increased (144). Using the same algorithm on a time series of hormone concentration alone has also shown higher approximate entropy of luteinizing hormone and testosterone concentrations in older males (145). Several diseases have been associated with decreased or increased complexity, compared to healthy subjects (142).

Porges introduced a model with an evolutionary approach (146–149). According to it, the autonomous system has three neural circuits: the myelinated vagus, the unmyelinated vagus, and the sympathetic–adrenal system (Table 1). All three circuits are involved in the regulation of physiological states. The sympathetic–adrenal part is mainly involved in mobilization related behaviors (fight and flight) whereas the vagal circuits are related to immobilization related behaviors. This approach can be discovered already in Cannon’s classical description (111), but Porges extended it to social behaviors. Initially, a system mainly aimed at homeostasis, ANS became during evolution increasingly important in the regulation of social behavior, according to this theory. The polyvagal theory proposes thus that the development of the mammalian ANS provides the neurophysiological substrates for the emotional experiences and affective processes that are major components of social behavior (147).

Thayer’s approach is based on the already mentioned complexity theories paradigm. He proposes to use HRV as a measure of the degree to which a system provides flexible, adaptive regulation of its component systems, with other words as a measure of the adaptivity of the brain–body system (150, 151). They describe their view as follows: “When processes mutually constrain one another, the system as a whole tends to oscillate spontaneously within a range of states. The various processes are balanced in their control of the entire system, and thus the system can respond flexibly to a range of inputs. However, such systems can also become unbalanced, and a particular process can come to dominate the system’s behavior, rendering it unresponsive to the normal range of inputs (…) A system, which is ‘locked in’ to a particular pattern, is dysregulated” (124). This approach has been developed further to a model called Neurovisceral Integration Model (124, 152, 153). Thayer and Lane proposed at the beginning that HRV might be a biomarker for emotion regulation, but they extended it eventually as a surrogate parameter for the more general top-down ability to self-regulation which is again coupled to the vagus nerve to the heart. As a general central nervous base for self-regulation, a Central Autonomic Network (CAN) has been defined (154). This network consists of parts of the prefrontal cortex (anterior cingulate, insula, orbitofrontal, and ventromedial cortex), limbic structure (amygdala and hypothalamus), and brain stem (periaqueductal gray matter, nucleus ambiguus, ventrolateral, and ventromedial medulla). These coupled structures and its oscillatory signal patterns are integrated into the nucleus of the solitary tract (NTS) and again coupled through efferent parts of the vagus nerve with organs outside the brain. The coupling is bidirectional such as peripheral oscillations in the heart, lung, but also in the immunological system (and others) can lead to changes in the CAN. According to the model, this regards especially the parasympathetic activity and should be associated with variations of the HF signal. This model has recently been expanded by defining eight levels of vagal control beginning with intra-cardiac control on the lowest level up to the highest levels where interactions between different parts of the prefrontal cortex shape the vagal tone over longer time periods (152). The model is sophisticated and based on new neuroscientific evidence. Especially on the network level, research is still needed. A recent investigation, for instance, reported no association between different individual levels of HF and general activity in the default mode network or the salience network in the human brain (155).

Besides these two models, other theories have been published. Grossman’s biological behavioral model focuses on the regulation of energy exchange by synchronizing respiratory and cardiovascular processes during metabolic and behavioral changes. Also there the main component is vagal signals, reflecting functional energy reserves as adaptive capacity (156). Lehrer and Gevirtz’s model is interested in the effects on slow pace breathing to increase vagal tone as a beneficent surrogate for health, mainly as a possible explanation for the effects of HRV biofeedback (157). McGraty and Childre have published a similar model within a broader context of physical and mental health (158).

Based on both approaches several studies have been published, mainly focusing on emotion regulation and executive control. Newer approaches also include sociodemographic co-variables (159), indicating that lower vagal components HRV mirror reduced adaptivity in self-regulation, supposed to be associated with lower sociodemographic status. This would indicate that HRV could be applicated as proxy for executive control and might be even used as measurement for therapeutical inventions. An association of lower sociodemocratic status and lower executive function, however, has been questioned in recent work (160).

An underreported issue is consequences of relatively higher HRV indices, based on the interest for the association between lower HRV and pathological conditions. Some evidence for that relatively higher HRV is associated with better mental and physical health has been reported. For instance, higher rMSSD is associated with better self-rated health (161).

A main issue in HRV research is that most studies report HRV parameters, but the main research question was not related to HRV. Frequently, relevant information or non-significant results are not reported. Relevant co-variables beyond age and gender, like Body Mass Index, physical status, or social background are often omitted, at least in the control groups. The well-documented association between immune status and different HRV results, for instance, is rarely reported, e.g., in the form of CRP values.

In the case of time series analysis, linear parameters have been established and are being used reliably. The non-linear part, however, is more difficult. A wide number of different forms of fractal measures, entropy measures or other non-linear indices have been introduced in the last two decades, but their relevance is still unclear. The relation between different HRV indices and the activity of various sympathetic and parasympathetic subsystems is also still unclear. In the last years, HRV parameters (especially HF) mirroring the function of the executive brain, in particular in the prefrontal cortex has been investigated. As pointed out by Holzman and Bridgett (162), a small, but a significant relation between top-down functions of the prefrontal cortex and HRV is now established, but more exact definitions and its clinical relevance remain to be shown.

Heart-rate variability is a fascinating observation and insight in its mechanisms is increasing. One main issue remains that interindividual differences are high and statistical effects are usually shown only on the group level. In particular regarding psychologic functioning, although an association between lower HRV and lower adaptivity is now established (162), the differences are frequently, although significant, very tiny—see as example recent work on HRV and depression (163), which makes its clinical application difficult.

From a systems science perspective, HRV measurement begins to be a sophisticated and relevant tool for both scientific and clinical insights.

GE has elaborated and written the review.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.