Heart rate variability in patients with cirrhosis: a systematic review and meta-analysis

Background. Cirrhosis is associated with abnormal autonomic function and regulation of cardiac rhythm. Measurement of heart rate variability (HRV) provides an accurate and non-invasive measurement of autonomic function as well as liver disease severity currently calculated using the MELD, UKELD, or Child–Pugh scores. This review assesses the methods employed for the measurement of HRV, and evaluates the alteration of HRV indices in cirrhosis, as well as their value in prognosis. Method. We undertook a systematic review using Medline, Embase and Pubmed databases in July 2020. Data were extracted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The risk of bias of the included studies was assessed by a modified version of the Newcastle–Ottawa Scale. The descriptive studies were analysed and the standardized mean differences of HRV indices were pooled. Results. Of the 247 studies generated from our search, 14 studies were included. One of the 14 studies was excluded from meta-analysis because it reported only the median of HRV indices. The studies included have a low risk of bias and include 583 patients with cirrhosis and 349 healthy controls. The HRV time and frequency domains were significantly lower in cirrhotic patients. Between-studies heterogeneity was high in most of the pooled studies (P < 0.05). Further, HRV indices predict survival independent of the severity of liver disease as assessed by MELD. Conclusion. HRV is decreased in patients with cirrhosis compared with healthy matched controls. HRV correlated with severity of liver disease and independently predicted survival. There was considerable variation in the methods used for HRV analysis, and this impedes interpretation and clinical applicability. Based on the data analysed, the standard deviation of inter-beat intervals (SDNN) and SDNN corrected for basal heart rate (cSDNN) are the most suitable indices for prognosis in patients with cirrhosis.


Introduction
Liver cirrhosis accounts for more than one million deaths annually worldwide, with numbers increasing year on year (Leon and McCambridge 2006;Schuppan and Afdhal 2008;Parra et al 2020). However, patients with cirrhosis have a range of conditions from early uncomplicated cirrhosis, which is asymptomatic, to decompensated cirrhosis where organ systems start to fail and patients present with many complications such as ascites, hepatic encephalopathy and variceal bleeding (D'Amico et al 2006; Kumar et al 2020). Once a patient starts to develop complications, various scoring systems including Model for End-Stage Liver Disease (MELD), MELD-Na, or United Kingdom Model for End-Stage Liver Disease (UKELD) are used to calculate prognosis and the need for liver transplant at the bedside or in the clinic. These scoring systems are widely available using apps on smartphones or web-based calculators. However, the scoring systems do not take into account the alteration in the autonomic nervous system (ANS) observed in cirrhosis (Barber et al 2007).
A simple and very useful tool to assess the state of the ANS is heart rate variability (HRV). HRV is the variation over time of the intervals between consecutive normal heartbeats (NN). Physiologically, instantaneous heart rate (HR) variation represents the capacity to adapt the HR to different internal and environmental circumstances and is modulated by the ANS (Rajendra Acharya et al 2006). Tsuji et al for the first time demonstrated the prognostic value of HRV analysis in the Framingham cohort study and reported that individuals with reduced HRV had increased risk for all-cause mortality (Tsuji et al 1994). HRV provides a noninvasive evaluation of autonomic regulation of the cardiac rhythm and indexes the interplay between the intrinsic cardiac rhythm and external regulatory controls (Vanderlei et al 2009). Importantly, with medical advances, it is becoming increasingly recognized that calculation of HRV from a continuous electrocardiogram (ECG) tracing provides additional and clinically useful information to clinicians on both the severity and prognosis of patients with cirrhosis (Bhogal et al 2018).
Although autonomic dysfunction and cirrhotic cardiomyopathy are well-established complications of cirrhosis (Fouad and Yehia 2014;Ferenci 2017), these are not assessed by MELD, UKELD or Child-Pugh. This is particularly important since recent reports suggest that HRV predicts survival in cirrhosis independently of MELD and Child-Pugh and therefore may provide additive information currently lacking in existing scoring systems (Mani et al 2009;Satti et al 2019;Bottaro et al 2020). However, for HRV to be of value in research and clinical practice, there needs to be standardization of processes that enables simple and accurate assessment of HRV. This includes standardization of the ECG recording techniques, duration of recording, methods, and clinical interpretation of HRV and also the availability of these at the point of care via mobile or web-based apps. The aim of this study was to analyse the methods used to record and report HRV in the literature, and assess HRV difference between patients with cirrhosis and controls.

Method
This systematic review follows the guidelines of Preferred Reporting in Systematic Reviews and Meta-Analysis (PRISMA) (Moher et al 2015). Embase, Medline and Pubmed databases were searched on the 8th of July 2020. An extensive search strategy using Medical Subject Headings (MeSH) terms was performed (figure S1 available online at stacks.iop.org/PMEA/42/055003/mmedia). Studies retrieved from the search were uploaded to Endnote and duplicates were removed. The duplicate-free studies were then uploaded to Rayyan software and title/abstract screening performed by two independent reviewers. Furthermore, a mini-systematic review was performed to assess the use of HRV indices as prognostic markers especially in the prediction of survival in patients with cirrhosis. Included papers were simultaneously screened and references searched for studies that performed survival analysis.

Inclusion and exclusion criteria
Only observational studies were included in the main review. Studies were considered eligible if any of the HRV time, frequency and non-linear indices were used in assessing autonomic cardiac control in cirrhosis. We excluded studies that did not include a control group; studies involving non-cirrhotic liver disease; studies involving pharmacological or non-pharmacological interventions known to affect HRV indices; and studies involving orthostatic tilting as the sole method of HRV assessment. For the mini-systematic review, studies were included if HRV indices of surviving and non-surviving patients were statistically compared irrespective of whether risk analysis was performed.

Data collection
Titles and abstracts were independently screened for potentially eligible studies and conflict about eligibility resolved through virtual meetings. The eligible articles were then analysed to identify studies meeting the inclusion criteria. All conflicts were resolved by a third reviewer.
Based on predefined criteria, the following data were extracted: the aims; summary of findings; sample and group sizes; study setting and country; etiologies of liver disease; length of ECG recording, and the equipment used; and methods of HRV analysis including data cleaning, analysed length, analysis software and indices calculated. The etiology of cirrhosis including alcohol, fatty, primary biliary cholangitis, viral and cryptogenic were also extracted (tables 1 and 2).
For the survival prediction, studies that used HRV indices for survival analysis in cirrhosis were included in the mini-systematic review. Where reported, the sample size, follow-up time, mortality, HRV indices analysed, HRV indices that predicts mortality independently of MELD and Child-Pugh scores, HRV indices of survivors and non-survivors and hazard or odds ratios were extracted from the studies (table 3). Child-Pugh /histological classification (Ates et al 2006) To assess using HRV autonomic dysfunction and its correlation with severity and 2-year survival in cirrhotic patients.
(1) HRV time-domain indices significantly reduced in cirrhotic patients compared to healthy subjects.
(2) HRV indices also significantly reduced in non-survivors versus survivors after 2 years of follow-up. To assess the effect on liver transplantation (LT)) on autonomic control of cardiac function in cirrhotic patients using HRV.
(1) HRV indices (SDNN and RMSSD) were reduced in cirrhotic patients compared with healthy subjects.
(2) LT corrected the reduced SDNN but RMSSD and LF/HF remained unchanged. To evaluate autonomic function in patients with liver cirrhosis using HRV and to evaluate the relationship with severity.

Italy
(1) HRV is significantly reduced in chronic liver disease.
(2) Reduced HRV is correlated with severity of liver disease. To determine if autonomic dysfunction is related to cerebral blood flow autoregulation in cirrhotic patients.
(1) Cerebral autoregulation of blood flow impaired in severe cases of liver cirrhosis.
(3) Severity of cirrhosis correlated with degree of autonomic dysfunction. (4) Loss of sympathetic innervation of cerebral vessels possibly linked with cerebral autoregulation dysfunction in cirrhotic patients.

Denmark
Outpatients To assess autonomic abnormalities in patients with liver cirrhosis using 123 I-metaiodobenzylguanidine (MIBG) myocardial scintigraphy and HRV.
Autonomic dysfunction is present in cirrhotic patients and can be assessed by MIBG myocardial scintigraphy and HRV.
(1) Psychological distress was correlated with increased serum aspartate aminotransferase (AST), and reduced autonomic control of the heart (HRV To analyse risk predictors of sudden cardiac death (SCD) related to autonomic dysfunction in alcohol-related cirrhotic patients.
(1) Patients with ARLD are susceptible to autonomic dysfunction (56% To assess the effect of portal blood flow volume and autonomic nervous function on abnormal gastric motility in cirrhotic patients. (1) Autonomic dysfunction correlated with abnormal gastric motility in cirrhotic patients. To assess sympathetic control of cardiac function in cirrhosis using MIBG scintigraphy and relate this to cardiovascular functions.
(1) Cirrhotic patients have significantly reduced HRV and baroreflex activity.
(2) Reduction in HRV and baroreflex correlated significantly with abnormal cardiac sympathetic nervous activity measured by catecholamine uptake by MIBG.

Denmark
Outpatients 10 (5) 10 (5) Alcohol=10 NR (Nagasako et al 2009) To assess autonomic dysfunction in non-alcoholic cirrhosis and the relationship with disturbed intestinal transit time, as well as severity and prognosis using HRV.
(2) HRV indices correlated significantly with the risk of hepatic encephalopathy in cirrhotic patients.
To assess the use of HRV as a marker of autonomic dysfunction and severity in cirrhotic patients.
(2) Aetiology of liver cirrhosis is linked with different types of autonomic dysfunctions.
(3) HRV parameters correlated significantly with severity of liver cirrhosis as assessed by Child-Pugh scores. To extend the Poincare plot by introducing a sequential lag in the correlation computation and to evaluate the relationship with the severity and survival of cirrhotic patients.
(1) Traditional SD1 and SD2 correlated strongly with severity of liver cirrhosis.
(2) Lagged SD1 and SD2 correlated significantly with liver disease severity. However, extended SD1 did not predict mortality independently of MELD.

Quality assessment
Two authors independently assessed the quality of the methods used in the included studies. A modified version of the Newcastle Ottawa Scale (NOS) was used (Negru et al 2015). The assessment comprised 3 domains that evaluated selection, comparability and outcome techniques employed in the included studies. The domains have 6 subdomains (questions) each of which is scored with a star. Lower scores (3 stars) were considered as having high risk of bias (table S1).

Data synthesis
A meta-analysis was computed to calculate the difference in HRV between patients with cirrhosis and healthy controls. Output was generated as forest plots of effect sizes of HRV indices between the groups using the Metan procedure in Stata/SE15. HRV indices reported as median and interquartile modes were transformed to mean and standard deviation according to (Luo et al 2018;Shi et al 2020), while indices presented as natural logarithms were transformed by finding exponents (e x ) accordingly. For studies where day and night indices were reported, the data of the day were used. The effect sizes in the form of standardized mean difference (SMD) of each of the reported indices were pooled using the Hedges' criteria (Borenstein et al 2011). Random-effect or fixed effect model were used depending on between-studies heterogeneity and effect sizes presented as SMD with 95% confidence intervals (CI) in HRV indices between the patients and healthy controls. The effect sizes were visualized as forest plots which include percentage weights and between-studies heterogeneity (I 2 Statistic, p-value). The I 2 statistic measures the degree of heterogeneity between pooled studies and can be putatively interpreted as low, moderate or high when values are 25%, 50% or 75% respectively (Higgins et al 2003 1 and 2).
Meta-analysis of the survival analysis was not performed because of the small number of studies found. Further, some of these studies originated from the same centre and possibly involved the same set of patients. Also, the indices reported varies between studies as well as the follow-up time.

Results
Our database search generated a total of 247 studies of which 67 were duplicates. Of the remaining 180 studies, a total of 30 studies were deemed potentially relevant according to the inclusion criteria. However, a full text review resulted in the exclusion of a further 16 papers to give 14 studies that fulfilled all criteria (figure 1). For the mini systematic review, a total of seven studies compared the HRV indices of patients with cirrhosis that survived and did not survive. Table 1 presents the general characteristics of included studies while table 2 contains the techniques used for ECG recording, HRV analysis and indices reported to be significantly different between the groups. The 14 included studies comprise a total of 583 patients and 349 healthy matched controls with sample size ranging between 20 and 180. All studies included are observational and conducted across 10 countries.

Description of included studies
The risk of bias as assessed by the modified NOS scale showed that most studies have low risk of bias with none of the 14 studies scoring3 (table S1). Overall, most studies reported a reduction in HRV indices except LF-HF ratio (LF:HF) which was reported to be both increased (Bhogal et al 2018;Kumar et al 2020) and decreased (Fouad and Yehia 2014) in cirrhosis. Table S2 contains definitions and units of the indices of HRV.
The mini-systematic review included seven studies involving a total of 437 patients. Of these, 104 (24%) patients did not survive to the end of follow-up (3-24 months). All studies observed significant differences in HRV indices between the survivors and non-survivors (table 3).

HRV time domains
A total of 13 studies observed significant differences in HRV time domains between patients with cirrhosis and matched controls. The time domains reported include standard deviation of NN intervals (SDNN), SDNN Index, standard deviation of the average NN intervals for each 5 min segment of a 24 h ECG recording (SDANN), root mean square of successive NN interval (RMSSD) and percentage of NN intervals that differ by 50% (pNN50). Significant effect sizes were observed for each of the pooled time domain indices (SDNN, SDANN, RMSSD and pNN50%) which were significantly higher in the healthy controls. Overall, the betweenstudies heterogeneities were significantly high in all the pooled HRV time domains (figures 2(a)-(d)).

SDNN
SDNN is the standard deviation of normal/non-ectopic RR intervals (NN) and is translated as the measure of the overall influence of the ANS on the variation of heart rhythm (Shaffer et al 2014;Shaffer and Ginsberg 2017). A total of nine studies reported significant reduction in SDNN in cirrhosis (Lazzeri et al 1997;Coelho et al 2001;Ates et al 2006;Frokjaer et al 2006;Mani et al 2009;Milovanovic et al 2009;Nagasako et al 2009;Ko et al 2013;Negru et al 2015). Eight of the nine studies presented SDNN as mean (±SD) while one study (Nagasako et al 2009) presented the median SDNN with no interquartile range (table S3). This study was excluded from the effect size computation. A very large effect size was observed with significantly higher SDNN in healthy control compared with patients with cirrhosis (SMD (95% CI)=3.41 (2.24, 4.58)); figure 2(a)]. This translates to noticeable dysregulation of the autonomic control of the cardiac rhythm in patients with cirrhosis.    SDANN (2b), RMSSD (2c) and pNN50 (2d) between patients with liver diseases and matched healthy controls. Hedges' G effect size estimates were calculated with 95% confidence interval and computed using random effect model. Continuous horizontal lines and diamonds width represents 95% CI, and the diamond centres and vertical red dotted lines indicate the pooled random effect sizes. (Shaffer et al 2014;Shaffer and Ginsberg 2017). Three studies reported a significant reduction in SDANN due to cirrhosis (table S5) (Lazzeri et al 1997;Ates et al 2006;Negru et al 2015). A significantly higher SDANN was observed in healthy controls compared with patients with cirrhosis with a very large difference between the groups (SMD (95% CI)=2.54 (0.81, 4.27); figure 2(b)).

RMSSD
The root mean square of successive NN intervals is the square root of the mean of squared differences in consecutive NN intervals. RMSSD has been linked with vagal influence on the heart rhythm and is used as an index of respiratory sinus arrhythmia (RSA) (Shaffer et al 2014;Shaffer and Ginsberg 2017). Overall, three studies reported significant alteration in RMSSD in the patients compared with healthy controls (table S6) (Lazzeri et al 1997;Ates et al 2006;Ko et al 2013). There was significantly higher RMSSD in the control group with a large difference between the groups (SMD (95% CI)=1.60 (0.73, 2.47); figure 2(c)). This can be interpreted as marked reduction in vagal control of the heart rhythm due to cirrhosis.

pNN50
The pNN50 is the percentage of the NN intervals that differ from each other by more than 50ms. The pNN50 indexes parasympathetic influence on heart rhythm and provides a comparatively less accurate assessment of RSA compared with the RMSSD (Shaffer et al 2014;Shaffer and Ginsberg 2017). Further, four of the pooled studies found significant reduction in pNN50 in cirrhosis compared with healthy control (Lazzeri et al 1997;Coelho et al 2001;Ates et al 2006;Nagasako et al 2009). One of the studies reported median pNN50 without the interquartile range and was not included on the analysis (table S7) (Nagasako et al 2009). Indeed, lower RMSSD was reported in the patients compared with the control groups with a very large effect size between the group (SMD (95%CI)=2.54 (1.21, 3.87); figure 2(d)).
HRV frequency domain HRV frequency domains represent the various frequency bands resulting from autoregressive (AR) or fast Fourier transformation (FFT) of the NN variations. Ten of the included studies reported significant difference in HRV frequency domains between patients with cirrhosis and healthy controls. Indices reported include total power (TP), high frequency (HF), low frequency (LF), very low frequency (VLF), and HF:LF. The betweenstudies heterogeneity as measured by chi-square was significantly high in pooled HF and LF (figures 3(b) and (c)). Thus, a random-effect model was employed. A fixed-effect model was used for pooling TP and VLF as the between studies heterogeneities were significantly low (figures 3(a) and (d)).
Total power (TP) TP represents the aggregate energy in all the frequency bands (ULF, VLF, LF, and HF) of an ECG recording (Shaffer et al 2014;Shaffer and Ginsberg 2017). Two studies reported significantly reduced TP in cirrhosis (Frokjaer et al 2006;Negru et al 2015). All reported TPs were analysed for 24 h ECG recordings and reported as natural logarithms in one of the included studies (Frokjaer et al 2006) (table S8). There was an overall higher TP observed in controls compared with the patients. A moderate effect size was observed between the group (SMD (95% CI)=0.62 (0.23, 1.02); figure 3(a)).

High frequency (HF)
High frequency is the power within the 0.15-0.4 Hz frequency band of an ECG recording. The HF represents the parasympathetic control of the heart rhythm and the variation linked with the respiratory sinus arrhythmia Of these, eight studies reported reduced HF while one study reported an increased HF in cirrhosis (Lazzeri et al 1997) (table S9). This study was not included in the data analysis because the model used does not correct for the difference in direction of effects (Deeks et al 2011). HF in healthy controls was observed to be higher compared with the patients, with an 'extremely large' effect size between the groups (SMD (95% CI)=4.36 (1.94, 6.77); figure 3(b)). This translates as an easily discernible dysregulation in vagal control of the heart rhythm in cirrhosis.

Low frequency (LF)
The LF domain represents the power within the 0.04-0.15 Hz frequency band of an ECG recording. The LF correlates strongly with baroreflex influence on the heart rhythm and is associated with both sympathetic and parasympathetic nervous controls (Shaffer et al 2014;Shaffer and Ginsberg 2017  , HF (3b), LF (3c) and VLF (3d) between patients with liver diseases and matched healthy controls. Hedges' G effect size estimates were calculated with 95% CI and computed using a random effect model. The width of the solid black diamonds represents 95% CI of the effect sizes of each of the pooled studies and the blue diamonds and vertical red dotted lines indicate the pooled random or fixed effect sizes. Frokjaer et al 2006;Mani et al 2009;Milovanovic et al 2009;Moller et al 2012;Ko et al 2013). LF power was observed to be significantly lower in cirrhosis compared with control with an extremely large difference observed in the effect size (SMD (95% CI) =5.49 (2.32, 8.67); figure 3(c)). Thus, there is a marked reduction in response of the heart rhythm to the the baroreflex loop which may be attributed to cirrhosis.
Very low frequency (VLF) VLF represents the power in the 0.0033-0.04 Hz frequency band of an ECG recording. While uncertainty exists in the physiological factors responsible for the VLF band, it has been strongly linked with the renin-angiotensin system, thermoregulation and endothelial factors (Shaffer et al 2014;Shaffer and Ginsberg 2017). Two studies reported significant difference in 24 h VLF between healthy controls and cirrhotic patients (table S11) (Frokjaer et al 2006;Negru et al 2015). The healthy control group were observed to have significantly higher VLF compared with patients with cirrhosis. The effect size for VLF between the group was moderate (SMD (95% CI)=0.73 (0.32, 1.13); figure 3(d)). This can be translated as a reasonably observable stronger response of the heart rhythm to the renin-angiotensin system, thermoregulation and endothelial factors in healthy controls compared with the patients with cirrhosis.
Low frequency-high frequency ratio (LF:HF) The ratio of LF power to HF power of an ECG recording, LF:HF, is traditionally translated as the measure of the sympatho-vagal balance because LF and HF had been hypothesized as measures of purely sympathetic and parasympathetic cardiac controls respectively (Shaffer et al 2014;Shaffer and Ginsberg 2017). However, this notion was challenged when it was observed that LF is influenced by both arms of the ANS (Billman 2013). A total of three studies reported significant differences in LF:HF between the patients and controls (Miyajima et al 2001;Iga et al 2003;Moller et al 2012). Of these, two studies reported increased LF:HF in patients with cirrhosis (Miyajima et al 2001;Iga et al 2003) and were pooled (table S12). However, no significant effect size exist between the groups (i.e. test that SMD=0: z=2.89, p=0.004).

HRV non-linear indices
A total of two studies reported significant differences in non-linear HRV indices between patients with cirrhosis and healthy controls (Mani et al 2009;Ko et al 2013). Four non-linear indices were reported including shortterm (SD1) and long-term (SD2) HRV extracted from the Poincare plot, sample entropy and scaling exponent (α) calculated using detrended fluctuation analysis (DFA). Short-term and long-term variability (SD1 and SD2) of Poincare plot as well as sample entropy were reported to be significantly reduced in cirrhosis in one study (table S13) (Mani et al 2009). The short-term scaling exponent (DFA α1) which indicates the fractal-like pattern of cardiac rhythm was also reported to be altered in cirrhosis compared with healthy controls in one study (Barber et al 2007). . All studies concluded that HRV indices were significantly different between survivors and non-survivors. Indeed, based on the hazard or odds ratios reported, increased DFA α2, SD2, cSDNN, SDNN and VLF were significantly correlated with increased survival (table 3). Of the seven studies, four reported that HRV indices including DFA α2, cSDNN (corrected SDNN), SD2, and SDNN may predict mortality in cirrhosis independent of MELD and/or Child-Pugh scores.

Discussion
We report here the effect of cirrhosis on autonomic cardiac regulation measured by indices of HRV. A significant difference in HRV was observed between patients with cirrhosis and healthy controls, albeit between-studies heterogeneities were high. To address the heterogeneity of included studies, we used a mostly random-effect model and pooled the SMD. In most cases, the HRV indices indicated autonomic dysfunction as shown previously (Amaral et al 2020). Some HRV indices (DFA α2, SD2, cSDNN and SDNN) also exhibited a significant correlation with severity of cirrhosis and survival of patients. More precisely, there was significant reduction in HRV time and frequency domain indices including SDNN, SDNN index, SDANN, RMSSD, pNN50 as well as TP, HF, LF and VLF in cirrhosis which correlated with disease severity. The relationship between cirrhosis and the ratio of LF:HF, which traditionally represents sympatho-vagal cardiac regulation, is not clear with one study reporting an increase (Moller et al 2012) and two studies reporting a decrease in cirrhosis (Miyajima et al 2001;Iga et al 2003).
Importantly, the mini-systematic review shows that some indices of HRV predict survival in cirrhosis independently of measures of severity (MELD score). This is consistent with a recent report whereby SDNN was shown to independently predict mortality in patients with decompensated cirrhosis. Further, Jansen et al also showed that SDNN was significantly reduced and correlated with increased plasma level of inflammatory biomarkers and severity of cirrhosis (Jansen et al 2019). Indeed, the association between cardiac autonomic dysregulation and systemic inflammation has been reported in cirrhosis (Mani et al 2009) as well as in other diseases (Lanza et al 2006;Lanza et al 2007). Similarly, the development of acute chronic liver failure (ACLF) has been extensively linked with systemic inflammation (Clària et al 2016;Mücke et al 2018;Lange 2019). Thus, it is suggested that a sudden reduction in HRV can be used to detect dynamic changes indicative of early acute decompensation in patients with cirrhosis (Jansen et al 2019). However, knowledge on the mechanism of cardiac autonomic dysfunction during systemic inflammation has only be reported in animal models and awaits further investigation in humans (Hajiasgharzadeh et al 2011;Gholami et al 2012;Haddadian et al 2013;Eftekhari et al 2020). Moreover, data are emerging to show that HRV correlates with sub-clinical hepatic encephalopathy (Mani et al 2009), and may therefore become an indirect means to identify patients most likely to have subclinical hepatic encephalopathy without the need for an electroencephalogram (Nabi and Bajaj 2014). Perhaps, HRV may provide a relatively simpler and portable yet effective assessment of covert hepatic encephalopathy.
Reduction in HRV is not unique to cirrhosis and has been reported in non-cirrhotic liver patients. For example, Keresztes et al reported significant reduction in HRV time and frequency domains in patients with primary biliary cholangitis compared with age-matched healthy controls. Indeed, 58% of the PBC patients studied also had autonomic dysfunction with abnormal cardiovascular reflex tests (Keresztes et al 2004). Further, patients with chronic hepatitis C infection have impaired autonomic function assessed by HRV time and frequency indices, and this correlates with the degree of liver injury as assessed by serum alanine aminotransferase levels (Osztovits et al 2009).
This study also observed that non-linear HRV indices were significantly impaired in cirrhosis. Short-and long-term HRV, measured by SD1 and SD2 respectively, and sample entropy were reduced in cirrhosis compared with healthy controls. Our findings substantiate the report by Bhogal et al in which SD2 as well as cSDNN were found to predict mortality independent of MELD in patients with cirrhosis (Bhogal et al 2019). The long-term fractal-like scaling exponent (DFA α2) was also lower in patients with cirrhosis predicting mortality independent of the measures of severity of liver failure (Mani et al 2009). Non-linear HRV indices are measures of unpredictability and provides an index of complexity in inter-beat intervals. Increased complexity may be interpreted physiologically as increased flexibility and inclination of the cardiac rhythm to respond to environmental changes and autonomic nervous control. Indeed, increased memory length of cardiac rhythm, which can be interpreted as reduced physiological controllability, has been reported in patients with liver cirrhosis compared to healthy controls (Shirazi et al 2013). Likewise, we recently showed that HR turbulence onset following premature ventricular contraction, a phenomenon that is linked with autonomic nervous control, is reduced and predicts survival in cirrhosis (Oyelade et al 2020). Put together, while the mechanistic link remains unclear, autonomic regulation of the cardiac rhythm is dysregulated in cirrhosis and may improve clinical diagnosis and prognosis of patients. Finally, while lower HRV complexity (sample entropy) has been reported in cirrhosis, its association with poor outcome has not been reported and should be a focus of future investigation.
Further, while HRV has been shown to be a potential candidate for clinical diagnosis and prognosis, it is important to be aware of dependency of HRV measures on experimental conditions such as patients' position (supine, seated, etc), time of the day (circadian rhythm) and how active the patient is. These factors have a significant effect on baroreflex and autonomic regulation of heart rhythm. This is particularly important in cirrhosis as baroreflex sensitivity is impaired and the patients have altered vascular compliance (Møller et al 2007). Thus, if HRV is going be utilized in clinical management of patients with cirrhosis, these various factors should be standardized. It is also well known that respiratory rhythm has a significant influence on HRV (particularly on short-term HRV indices) since cirrhosis is associated with respiratory complications such as hepatopulmonary syndrome (Wessel et al 2009;Grilo-Bensusan and Pascasio-Acevedo 2016). It is essential to also consider the influences of respiratory rhythm on HR dynamics in patients with cirrhosis in future investigations.
This study has several limitations. Because of the difference in the time of ECG recording in the studies, it was not possible to assess the role of circadian rhythm in HRV. Further, few studies reported non-linear indices of HRV, and thus data could not be pooled to assess the differences or the survival analysis because of the variability in the follow-up time. Finally, as only 5% of the generated studies were included, the result of this systematic review may be skewed by study location, selection and publication biases.
We conclude that HRV analysis has the potential to be applied in both clinical and research settings for the evaluation of autonomic dysfunction in patients with cirrhosis and may be of value in the early detection of subclinical deterioration such as early covert hepatic encephalopathy or acute chronic liver decompensation.
Further, HRV is an independent predictor of the outcome of cirrhosis and may improve the predictive power of MELD and Child-Pugh. Finally, despite the potential, variation in techniques of HRV measurement remains the main barrier to useful interpretation and applicability. Indeed, to generate robust data, there needs to be standardization of techniques including ECG recording and HRV measurement in the future. We hope that this paper provides an impetus to agreed standardization of methodology to enable this field to develop.