Introduction

Remarkable progress has been made in the basic understanding of the systemic host response to infection, with detailed characterization of the whole organ, endocrine, cellular, endothelial, intracellular, and molecular mechanisms that all interact in a complex orchestra of interdependent pathways. In addition, population studies have led to dramatic improvements in care in a variety of respects—in particular, with evidence demonstrating the critical importance of early appropriate antimicrobial coverage, early goal-directed therapy, protocoled use of bundles of care, and more. Despite the significant progress made in basic science understanding and more effective management strategies, clinicians still use traditional means to diagnose infection and sepsis on the basis of the Surviving Sepsis Campaign [1], which defines sepsis, severe sepsis, and septic shock by monitoring of absolute values in vital signs, with delayed confirmation based on microbial cultures and observed clinical course. Despite enormous quantities of data being harvested at the bedside, using continuous vital sign monitoring, the vast majority of data are discarded, focusing only on intermittent point-in-time vital sign measurement.

Considering the involvement of inflammatory, endocrine, coagulation, metabolic, and end-organ interactions, pathways, and feedback loops, it is clear that the host response to severe infection or injury is a complex nonlinear system [2, 3]. Identifying the host response to severe insult as a complex system helps explain why chaotic dynamics and cascade behavior may lead to unexpected rapid deterioration in patients with infection and, conversely, to unexpected clinical improvement with no identifiable cause. If critical illness is characterized by a complex systemic response, continuous monitoring over time of the system as a whole allows tracking the changes (i.e., trajectory) in the system state. Since the temporal variations of a biological system are produced from the integrity and complexity of the whole system, continuous monitoring of variability may offer a way to monitor the whole system and its changes over time [2].

Over 3 decades, research in mathematical physics, nonlinear dynamics, and biomedical engineering has led to breakthroughs in the techniques and understanding of how to characterize patterns of variation over an interval-in-time. The scientific study of both the degree and character of variation of a time series is referred to as variability analysis. Variability aims to provide information that adds incremental value to standard absolute value vital sign assessment. In the following review and analysis, we will discuss the relevance of heart rate, respiratory rate, temperature, and glucose variability in the context of infection. The objectives of this review and analysis are to explore the current data available with regard to variability and its application to sepsis and to discuss how variability may broadly offer value to the methods by which clinicians diagnose, manage, and treat sepsis.

Background

Severe sepsis and septic shock are the most common causes of mortality in critically ill patients, accounting for 10 % of intensive care unit admissions [4] and 2.9 % of all hospital admissions [5].The mortality rate of severe sepsis syndrome—namely, infection leading to hypotension and organ failure—admitted to ICU has remained at approximately 50 % over the last 2 decades [6]. The average annual cost for treating severe sepsis amounts to $16.7 billion in the U.S. [5]. In the case of severe sepsis or septic shock, early diagnosis with aggressive resuscitation has emerged as a critical means for reducing mortality, using goal-directed rapid expansion of plasma volume and optimizing perfusion and oxygen delivery to the tissues [7]. A large randomized controlled trial demonstrated that an early goal-directed therapy aimed at maintaining adequate oxygen delivery and tissue perfusion reduced subsequent organ dysfunction and improved hospital survival in patients presenting to the emergency room with sepsis [8]. The benefit of this goal-directed resuscitation and restoration of perfusion appears to exist in the early hours of sepsis, in contrast to lack of benefit in established sepsis or organ failure [8]. A comprehensive evaluation of early goal-directed therapy for the treatment of sepsis and septic shock revealed that this treatment strategy resulted in both improved patient outcome and cost effectiveness in treating sepsis [9]. As well, effective and early antibiotic therapy has been shown to be critical in reducing mortality secondary to sepsis [10]. Given the proven benefit of early resuscitation and antibiotics in sepsis, there is an additional imperative to develop means for diagnosing infection earlier, with the potential to save lives.

Currently, the diagnosis of sepsis is made by clinical suspicion and by monitoring absolute values of vital signs and laboratory values. In the setting of infection, once abnormal thresholds are reached in the deterioration of vital signs (e.g., fever), the diagnosis is made. The diagnosis of bacterial infection is often made on a gestalt of individually nonspecific clinical signs and symptoms. Given that the criteria for diagnosing infection are nonspecific, the timing of diagnosis is imprecise and may lead to delay in the initiation of antibiotic therapy. However, despite the fact that the diagnosis of infection is not always precise; the Surviving Sepsis Campaign has laid out criteria that define sepsis, severe sepsis, and septic shock [1]. Specifically, sepsis is defined by evidence or suspicion of infection combined with an altered systemic host response, as captured by the Systemic Inflammatory Response Syndrome (SIRS) criteria (two or more of elevated heart rate or respiratory rate, elevated or depressed temperature, or white blood cell count). Severe sepsis is defined as sepsis with resultant acute organ failure, and finally, septic shock is defined as severe sepsis with hypotension not correctable with fluid resuscitation [11]. While these definitions have greatly added to the epidemiologic understanding of sepsis, the nonspecific signs of infection make it difficult to establish early diagnosis of infection in an individual patient, determine the risk of deterioration, and gauge whether he or she is responding favorably to treatment. Thus, current methods of diagnosis of infection and managing patients with existing sepsis are ripe for improvement.

Variability Analysis

Variability analysis is the science that characterizes the degree and character of patterns of variation of a time series occurring as a succession of events in time [12]. For instance, the analysis of the changes of the time series of R–R intervals (time interval between two successive R waves of an electrocardiogram) has been termed heart rate variability (HRV). The research paradigm we have pursued with respect to the interpretation of HRV is complex system science. The fundamental property of a complex system is that, as a whole, it produces behaviors that emerge from the integrity of the whole and disappear when the system is broken up into its component parts. This phenomenon arises from the intrinsic nonlinear nature of the interactions (e.g., feedback loops and cascade behavior) between the system components [13], as well as from the dissipative nature of complex systems; that is, they break down energy gradients, produce entropy, and create order [14•]. The nonlinear interactions and potential for cascade-like events enable the system to show large, abrupt changes to small perturbations (i.e., chaotic behavior) and/or variations that are consistent at different scales (i.e., fractal temporal variation). The understanding of complex systems and the patterns of variation produced from them represents an exciting and still incomplete area of active investigation.

In parallel with seeking understanding of complex variation, researchers have developed a variety of measures to quantify them [15•, 16•]. Those measures of variability are divided into domains, with each domain representing a particular type of information that a measure is able to extract. The hypothesis underlying the classification of the measures into domains is that each type of information (be it statistical, frequential, entropic, etc.) can be associated to a given physiological phenomenon. The techniques for variability analysis remain an ongoing area of investigation.

The third area of research regarding variability has focused on the clinical association between altered variability and illness states, from aging to cardiovascular disease to critical illness. The subject of this review and analysis is the clinical research that has documented altered variability in association with sepsis—namely, the presence of infection and an altered host response to that infection. In this article, we will focus on the areas of greatest interest and accessibility in heart rate, respiratory rate, temperature, and glucose variability Table 1.

Variability and Sepsis

Heart Rate Variability and Sepsis

As previously described, HRV represents the study of fluctuations within the R–R interval time series. Given the accessibility of recording an electrocardiogram in a clinical context and the relatively straightforward processing required to identify the R peaks, HRV has been widely studied for several years [17]. HRV has found several applications in a variety of clinical contexts, ranging from detection of myocardial infarction [1820], diabetic neuropathy [21], cardiac transplantation, myocardial dysfunction, tetraplegia [17], chronic heart failure [22, 23], autonomous nervous system dysfunction in Parkinson’s disease [24], and several others [25, 26•]. It is commonly believed that HRV and, more specifically, a set of frequential measures, such as balance of low and high frequencies (LFs and HFs) of the power spectrum (for more details on these measures, refer to [16•]), reflect modulation from the sympathetic and parasympathetic activity of the autonomic nervous system. Given the extensive involvement of the cardiovascular system during sepsis (including altered cardiac output, contractility, sympathetic activity, etc.), the relevance of HRV to monitoring development of infection associated with altered systemic host response (i.e., sepsis) has been carefully studied. The following discussion highlights recent literature demonstrating the alterations of HRV presenting in association with sepsis, focusing on how monitoring HRV may prove useful clinically

We have previously reviewed the association between HRV and infection, highlighting how altered HRV is present with the presence of infection and correlates with illness severity [27]. Here, we focus on studies that provide clinically useful information at the bedside based on monitoring HRV that may be utilized to alter patient care. In 2002, Barnaby et al. used continuous heart rate monitoring in a pilot study of 15 patients presenting to the ED with sepsis [28]. Altered HRV (specifically, LFnu values less than 0.5 and LF/HF ratios below 1.0) identified patients with greater risk of developing shock, organ failure, or death. Furthermore, a 2007 study by Chen and Kuo monitored HRV in 81 patients presenting to the ED with sepsis in order to predict impending septic shock [29]. This study also showed that septic patients who subsequently developed shock (defined by onset of hypotension, altered LOC, or oliguria) had altered HRV on their initial 10-min EKG analysis (decreased LF and LF/HF power). Recently, a prospective observational study by Günther et al. showed the utility of HRV as a marker for subacute infection presenting following a stroke [30]. The study involved 43 patients with acute ischemic stroke, 9 of whom developed acute infection and 12 of whom developed subacute infection. HRV analysis was performed on normal-to-normal R–R interval time series of 3 h (Holter recordings) during daytime and nighttime (to account for circadian fluctuations), at two specific moments: (1) acute period (within 48 h after admission) and (2) subacute period (4 ± 1 days after admission). Using indexes from the statistical (standard deviation and root mean square of square sum of adjacent differences) and the frequential domain (power at very low frequencies, LFs, and HFs), they identified that a decreased normalized power at the LFs (LFnorm), a decreased LF to HF ratio (LF/HF), and an increased normalized power at the HFs (HFnorm) were signs of subacute infection. We remind the reader that those measures are extracted by estimating the level of power of an R–R interval time series in well-defined frequency ranges (LF, 0.04–0.15 Hz; HF, 0.15–0.4 Hz; for details, refer to Table 2 of [17]). Trying to relate variability with sepsis pathophysiology, Papaioannou et al. characterized the association between HRV and inflammatory biomarkers in 45 septic patients [31]. The study was an empirical evaluation of the levels of correlation between measures of HRV and a set of clinical parameters extracted from the serum (C-reactive protein, interleukin 6, interleukin 10) and Sequential Organ Failure Assessment (SOFA) scores. The objective was to create a link between variability and the pathophysiology of sepsis and was based on the underlying assumption that loss of variability should be associated with the failure of ANS-derived inflammation-suppressing mechanisms. For each patient, electrocardiogram traces of 10 min and, subsequently, a 128-s artifact-free R–R interval time series were extracted daily. A strong linear correlation between the time evolution of the levels of biomarkers, the SOFA scores, and a small set of measures of variability from the statistical, frequential, and geometric domains (standard deviation, LF, HF, LH/HF, and Poincaré plot SD1/SD2) was identified, providing additional proof of the utility of HRV in sepsis monitoring and identification. From our group, Ahmad et al. in 2009 studied HRV analysis over many days during the onset and resolution of sepsis in a population of 17 adult bone marrow transplant (BMT) patients. Continuous Holter ECG data were collected for all participants from 24 h prior to BMT until resolution of neutropenia or withdrawal from the study [32]. Individualized percentage change in variability was calculated with respect to the baseline variability (first 24 h). Out of the 14 patients who developed infection, 12 experienced a significant (25 %) drop in HRV prior to the development of sepsis, on average, 24 h prior. None of the 3 patients without sepsis demonstrated a loss in HRV variability. These studies support the theory that HRV monitoring might be useful for helping physicians both identify and predict sepsis and its severity.

Table 1 Variability and Sepsis

Leading the way in terms of bedside clinical application with the highest level of evidence in this domain has been the work done in the infant population by Moorman, Griffin, and colleagues. By combining standard deviation, sample entropy, and clinical factors through a logistic regression model, they created the heart rate characteristic (HRC) model [33, 34]. Following derivation and validation of thresholds of HRC that predicted an increased likelihood of deterioration in the subsequent 24 h, Moorman et al. ran a large multicenter randomized controlled clinical trial involving over 3,000 neonates with very low birth weight [35•]. Nine intensive care units were provided with monitors displaying an index showing the fold-increase in the risk of developing neonatal sepsis in the subsequent 24 h. The study showed that by displaying the HRC values at the bedside, a reduction from 10.2 % to 8.1 % in the in-hospital mortality rate was obtained (p = .04). This reduction was statistically significant and clinically relevant, and particularly noteworthy given the fact that the individual clinician’s response was left up to the clinician’s judgment.

The collection of evidence in the recent literature suggests that HRV has something to offer for the care for septic patients, albeit a greater level of evidence is required. This is already happening for neonatal sepsis, while several principal challenges remain within the domain of adult sepsis: (1) First, standardization of the methodology for performing variability analysis will enable comparison between centers and studies; (2) second, large multicenter studies are essential to deriving and validating the thresholds of HRV that are associated with predefined clinical outcomes such as septic shock, as described above in several pilot studies; (3) the sensitivity, specificity, and positive and negative predictive values of HRV have yet to be determined; and (4) the cost effectiveness and overall clinical impact of variability-directed clinical decision support would need to be formally evaluated in RCTs. In addition, (5) because several measures of variability have been found able to track sepsis development, there is the need for a way to merge all the clinically relevant information and eliminate the redundant; and (6) last, the physiologic understanding of altered HRV remains incomplete. For example, the impact of cofounding factors on HRV requires greater understanding. For example, many in the patient population in which HRV may prove useful are on inotropes and vasopressors in the ICU. It would be important to examine the influence these agents have on HRV in the setting of sepsis. In addition, as was previously highlighted, HRV has been studied in many different comorbidities, many of which patients presenting with sepsis may have as underlying medical conditions. It would be important in future studies to consider how underlying medical comorbidities may influence HRV in people with sepsis. Thus, numerous challenges remain, in order to implement HRV in the care of adult patients with sepsis.

Respiratory Rate Variability and Sepsis

Respiratory rate variability (RRV) represents the study of fluctuations in the time elapsed between two successive breaths—namely, the interbreath interval (IBI) time series of normal breaths. The IBI time series can be extracted from flow measurements, CO2 capnography, or plethysmographic methods [36]. Contrary to HRV, there is no theory attempting to explain the meaning behind variations in the patterns of breathing. Indeed, only recently has the use of RRV been evaluated, mainly for the assessment of weaning from mechanical ventilation [3739] and extubation outcome prediction [40].

Recently, a scoring system to account for the variety of therapeutic responses in septic patients—namely, PIRO (predisposition, infection, response, organ failure)—has been proposed [41]. Howell et al. used three independent, observational, prospective cohorts: a derivation with 2,132 patients and two validations (one internal, the other external), with 4,618 and 1,004 patients, respectively. Although not related to RRV, by using a multivariate logistic regression model, this system highlighted the relevance of the respiratory system at a variety of levels: (1) predisposition level (chronic obstructive pulmonary disease), (2) infection level (pneumonia), (3) response level (tachypnea), and (4) organ dysfunction level (respiratory). To the best of our knowledge, no one has tried to use RRV for sepsis monitoring and/or identification. Given that sepsis develops to shock and respiratory failure, it is likely that there is a progressive loss of RRV. Theoretical arguments, such as the modeling of the host response to infection as a complex system, together with empirical evidence, such as the PIRO system, provide enough reasons to justify the exploration of the utility of RRV for this application.

As a former step required for a proper application of RRV, we need to highlight the fact that more science has to be put on the extraction and the assessment of the quality of the IBI time series. Indeed, there are no tools currently available to make sure that variability is computed on data of high quality. Depending on the measure of variability, low-quality data could produce particularly poor results, shadowing the potential clinical utility of RRV. In addition, once the methodology is in place to study the utility of RRV, confounding factors will also need to be determined. For example, in the setting of sepsis in the ICU, mechanical ventilation would be a confounding variable in studying RRV. Furthermore, ventilator settings, sedation, and narcotics would need to be examined to determine their impact on RRV.

Temperature Variability and Sepsis

Temperature signals do not present in recurring measurable cycles , such as a QRS complex in the electrocardiogram or the breath wave into a CO2 capnogram; therefore, variability analysis is directly performed on the temperature time series, sampled regularly in time, without the extraction of intermediate waveform time series (such as R–R interval or IBI time series). Since the former work of Varela et al. [42], temperature measurements for variability analysis have been usually performed through thermistors sensors with a resolution of 0.05 °C, placed at either the right or left hypochondrium. The interest behind temperature variability (TV) relies on the evidence that temperature fluctuations arise from complex regulatory processes (thermogenesis, thermoregulation) and the hypothesis that their interaction with the environment may be affected by the pathophysiology of certain diseases. In support of this hypothesis, changes in TV has been found to be related to aging in healthy subjects [42] and to the degree of illness and survival in critically ill patients [4345].

A study published by the Varela et al. (2006) [44] was designed to investigate the degree to which temperature curve complexity is capable of predicting patient survival or death and how it compares with a conventional scoring system that assesses severity of organ dysfunction (SOFA score). This study consisted of 50 patients admitted to ICU suffering from multiple organ failure. The study measured temperature readings every 10 min, and this series of data was used to calculate the approximate entropy and the detrended fluctuation analysis value for each successive period. The minimum level of complexity and the mean complexity value for the ICU admission were calculated for each patient and compared with a SOFA score. The results of this study confirm that in critically ill patients, the complexity of the temperature curve is inversely related to the patient’s clinical status. Essentially, in the case of the body’s thermoregulatory system, a poor clinical condition is associated with decreased TV, which can be measured using complexity analysis. Furthermore, the predictive power of complexity analysis to forecast the clinical outcome was found to be similar to that of the more traditional method of using a SOFA score. By calculating the area under the receiver operating curve comparison, no statistically significant difference between the areas under the curve of the SOFA score and those values (DFA and ApEn) representing variability was found. Complexity can be monitored to predict outcomes with a similar result as using SOFA scores, thus supporting TV as a monitor of outcome. Subsequently, Cuesta et al. (2007) [45] discussed the possibility that the correlation between body temperature and a patient’s condition may be missed if it is measured only at intervals in time. This study looked at measuring temperature at a higher frequency and monitoring changes in patterns; thus, it looked at the clinical implications of the body’s temperature complexity curve. It was found that severely ill patients seem to thermoregulate poorly, observed by the blunting of temperature variability. The results showed a statistically significant difference between the regularity of temperature signals from patients that survived and patients that did not. This study was consistent with previous studies that showed a relationship between the loss of complexity of temperature time series and the clinical status of patients. These findings supported the hypothesis that continuous temperature monitoring and use of temperature curve complexity may be a tool that can help to predict patient outcome

Absolute value of temperature is the classic physiological variable on which the diagnosis of infection is made. It has been used as a diagnostic tool for thousands of years, interpreted usually in a binary fashion; if the patient has reached a predefined threshold, the patient is febrile. Typically, this temperature cutoff that delineates a “fever” may be a temperature equal to or greater than 38.0 °C, as laid out by the SIRS criteria, or greater than 38.5 °C, as is commonly practiced by many. However, TV—namely, temperature fluctuations measured over intervals in time—may contain additional information. Papaioannou et al. recently investigated the relationship between TV and different degrees of systemic inflammation [46]. More specifically, the goal of their study was to track dynamic changes of continuously monitored temperature to see whether it was indicative of disregulated hemostasis of a complex thermoregulatory system during inflammation. This pilot study enrolled 22 patients suffering from systemic inflammation during a suspected ICU-acquired infection. The study monitored temperature by using a thermistor sensor that sampled body temperature every 10 s during the first 24 h of an inflammatory state. Of these 22 patients, 5 developed SIRS, 10 developed sepsis, and 7 developed septic shock. A variety of entropic measures (discrete and continuous Wavelet entropy at different scales, sample entropy, and multiscale entropy) were found different between patients with SIRS and those with septic shock. By using unsupervised and supervised machine-learning techniques, the authors provided further evidence of the separability between SIRS and higher degrees of infection. The differentiation between SIRS and sepsis and between sepsis and septic shock seemed possible but produced less robust results. It is noteworthy that the authors concluded the article by providing an extensive discussion about the possible relationship between TV and systemic inflammation pathophysiology. These include the following: (1) alteration of the frequential components of skin temperature fluctuations associated to sympathetic nervous system activity; (2) alteration of blood flow due to the release of different molecules from the endothelium, thereby relating to a change in local metabolic activity (such as the release of NO synthase, producing a reduction in smooth muscle cell oscillation); and (3) alterations of the circadian rhythms of free tumor-necrosis factor-α (TNF-α). The authors highlighted a variety of limitations in their study. Notably, the major one is that different treatments were provided for subjects in different classes of infection. Indeed, the differentiation between illness and treatment is always challenging—in particular, when it involves observational variability research. Patients who developed severe sepsis and septic shock received noradrenaline (a vasopressor) to restore the appropriate blood pressure, thereby altering the thermoregulatory action of the cardiovascular system. Furthermore, the authors noticed the relevance of assessing the potential sensitivity of TV to different anatomical areas for temperature measurement. The studies on TV published so far have always used a temperature sensor placed in the right or left hypochondrium; however, no proof was provided regarding the assumption that different loci have identical TVs. The answer to this characterization may provide valuable insights into TV, which may produce beneficial effects on its clinical application. Similarly, this characterization should be done taking into account the environmental conditions affecting skin temperature, such as humidity and temperature. In conclusion, further studies on a larger scale are needed to evaluate these initial findings and to establish a way to translate them into clinical practice.

Glucose Variability and Sepsis

Glucose variability (GV) represents the fluctuations of the concentration of glucose in the blood. Measured through glucometers, in a recent review, GV has been confirmed to be a key component for the treatment of diabetes, together with other pathologies, peripheral neuropathy, retinopathy, acute myocardial infarction, and severe hypoglycemia [47]. Related to the acutely ill, augmented GV proved to be strongly related to mortality in critically ill patients in several large population studies [4851]. The principal proposed mechanism for the physiology explaining the association between GV and mortality is oxidative stress related to periods of elevated glucose [47]. In vitro studies [52, 53] show that intermittent high glucose levels stimulate reactive oxygen species that can induce apoptosis. However, these findings have not been reproduced in humans [47]. Unlike heart rate and TV, discussed above, GV has not been looked at as a monitoring tool but, rather, as a controllable variable for therapeutic intervention.

Control of blood sugar levels has been well established as part of sepsis management. As per the Surviving Sepsis Campaign, patients with severe sepsis and hyperglycemia should receive IV insulin therapy to prevent hyperglycemia. On the basis of data from the NICE-SUGAR trial, the Surviving Sepsis Campaign recommends a target threshold less than <180 mg/dL for critically ill patients [54]. An extensive review of the relationship between glucose physiology and the inflammatory response is provided by Collier et al. [55]. Given the clear benefit of preventing hyperglycemia, recent studies have looked at the role of GV in septic patients. In a recent review article [47], increased GV was shown to be associated with mortality in patients admitted to hospital and, specifically, in critically ill patients. There have been numerous studies investigating the role of glucose in the critically ill, but relatively little has been published about GV and sepsis specifically. In 2008, Ali et al. completed a retrospective cohort study of 1,246 patients admitted with sepsis (as per discharge diagnosis) [56]. GV was examined using standard deviation and two measures of variability particularly designed for GV—namely, the glycemic lability index (GLI) and the mean amplitude of glycemic excursion. Of these three methods, the study determined that GLI had a significantly better discrimination of in-hospital mortality. The relationship between GV and mortality in septic patients was found independently of several factors, such as hypoglycemia, the presence of diabetes, and organ failure. This study showed that variability was associated with increased hospital mortality and was significantly increased in those at a lower average glucose level, concluding that GV is an independent factor with increased hospital mortality.

GV is therefore highly different from the other types of variability previously discussed. Glucose is challenging to measure precisely and continuously; however, GV, being based on the fluctuations of a metabolite, represents a potentially controllable therapeutic target. Several articles have discussed the methods and the expected outcome for this type of control [57, 58].

Discussion

While evidence to date has largely been derived from single-center studies, variability has nonetheless shown the potential to be a tool that adds value to the diagnosis and management of patients with sepsis. Continuous HRV monitoring may assist in early detection and prognostication of infection. With the greatest foundation of evidence to date, HRV monitoring in neonates appears to save lives simply by alerting physicians that something may be wrong. Along with TV, monitoring HRV and RRV to track severity of illness of infection over time could assist in patient management. As for treatment, the impact of variability on therapeutic interventions for sepsis still remains hypothetical. The potential for variability monitoring to provide individualized feedback of response to intervention (i.e., favorable–continue intervention or unfavorable–discontinue intervention) is perhaps the most exciting potential clinical application, yet remains furthest from evaluation in our patients. Animal models will help determine how variability is altered with infection and interventions and help evaluate whether a pathophysiologic response may be converted to a physiologic response by tuning variability. Further research is required to explore different aspects of variability, and significant work needs to be done to put these findings into practice.

First, given the assumption that the body is a complex system, we believe that there is high interest in trying to characterize the connectivity of different physiological functions (e.g., cardio-respiratory), as highlighted by Cerutti et al. for the cardiovascular system [26•]. To date, there is a lack of evidence showing how the different elements of variability relate to each other [16•]. Each measure of variability captures certain properties of a physiological system under study, and there is the need to use several measures of variability, because each one provides a single piece of information that may prove useful. While there is need to eliminate any redundancy of variability metrics, it is simultaneously evident that more research is needed to develop integrated measures of comprehensive variability—that is, monitoring HRV, RRV, TV, and GV concurrently in order to monitor the system as a whole. We hypothesize, simply by extension, that it would be useful to combine information from multiple organ system variability to monitor infection and its severity and to avoid the harms of certain therapies, which may reduce HRV, TV, or RRV or augment GV. The heterogeneous information brought by HRV, RRV, TV, and GV would likely be enhanced by assessing the information shared between their underlying signals. It represents an in-depth integration over both time and space. For example, characterizing the synchrony between heart beat and respiratory rate may provide additional clinically valuable information. Similarly, the inclusion of additional sources of variability, such as QT time variability and T wave shape variability, may provide additional valuable information. With this in mind, the concurrent monitoring of multiple organ systems through variability analysis would allow for the comparison of different trends in HRV, RRV, TV, and GV.

In contrast to amalgamating multiple organ system variability, further research is needed to determine whether trends in individual organ variability may provide specific clinical information on the organ system affected. For example, if RRV is the most affected, would this indicate a respiratory infection? This is tied to better understanding of what variability is telling us about physiology. While not directly clinically impactful, the understanding of the physiology and pathophysiology of altered variability is critical to clinical application.

Third, whereas we have discussed the clinical relevance of variability as found in various studies, there is a lack of literature on whether or not the theories hold true in large multicenter trials and what their applicability in real-life ICU settings is. As outlined in this review, many small studies indicate that variability has the potential to provide assistance with early detection and prognostication of sepsis. However, larger trials are needed to validate these findings and to study their relevance on a broader scale. It is possible that a key aspect to implementing variability at the bedside will be to integrate all the clinically valuable information, useful for a specific application, into a unique measure—namely, a composite measure of variability. Regardless, for variability measures to be integrated into clinical decision making, large multicenter derivation and validation studies followed by randomized controlled trials will be needed to determine the thresholds of variability that predict clinical outcomes, apply them, and determine their impact.

Conclusion

The focus on variation of vital signs, in contrast to the traditional focus on absolute values, represents a complementary means of monitoring patients with sepsis. Variability shows potential as a means for detecting infection in a more timely manner than current practice allows for today. Specifically, to date, HRV has been shown to decrease prior to the onset of severe infection and sepsis, thus providing a novel method for detecting infection and sepsis and predicting outcome. Along these lines, other forms of variability (RRV, TV) could foreseeably follow suit and provide physicians with additional information about patients’ clinical picture, but data are lacking at present. Ultimately, the end goal of studying variability would be to develop a monitoring system to measure real-time fluctuation in variability that would serve as a tool for physicians to facilitate a diagnosis of sepsis in order to implement early goal-directed and variability-directed therapy. Despite these promising findings in the literature, further research is still needed for this to be used as a standard technique in clinical practice. If this can be accomplished, variability monitoring offers promise to enhance future patient care.