Use of heart rate variability to predict hospital length of stay for COVID-19 patients: A prospective observational study.

BACKGROUND: As the COVID-19 pandemic continues, determining hospital demands has become a vital priority. Heart rate variability (HRV) has been linked to both the presence of viral infection and its severity. We investigate the possibility of using HRV parameters in comparison to other clinical parameters for predicting the hospital length of stay (LOS) for COVID-19 patients.
METHODS: This was a population-based cohort study. Measurements were performed in a specialized hospital for respiratory disease, dedicated to COVID-19. Patients were polymerase chain reaction positive for COVID-19 and on their 1st day of admission. Heart period, respiratory sinus arrhythmia (RSA), low frequency (LF) HRV, and vagal efficiency were calculated from electrocardiogram signals. This study investigated the correlation of HRV, demographic, and laboratory parameters with hospital LOS.
RESULTS: Forty-one participants were recruited, with a significant relationship, observed between hospital LOS and some demographic and clinical parameters such as lymphocyte count, age, and oxygen saturation of arterial blood. There was a negative relationship between LF and hospital LOS (r = -0.53, 95% confidence interval: -0.73, -0.24). Higher vagal efficiency predicted shorter hospital LOS in patients younger than 40 years of age (19.27% shorter hospital LOS was associated with a one SD higher value of VE, P = 0.007).
CONCLUSION: HRV measurement is a non-invasive, inexpensive, and scalable procedure that produces several metrics, some of which are useful for predicting hospital LOS and managing treatment resources during COVID-19 pandemic. Copyright:

INTRODUCTION

A new strain of beta coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is the cause of a new respiratory syndrome identified as coronavirus disease 2019 (COVID-19).[1] According to a report recently released by the WHO-China Joint Mission, 19.9% of people infected with SARS-CoV-2 have severe or critical disease,[1] typically requiring hospitalization. Since longer hospital length of stay (LOS) may correlate with greater resource utilization, inexpensive noninvasive, and effective tools are needed to early identify such patients.[23]

The cardiovascular system mediates exchange of respiratory gases between the lungs and body cells.[3] Heart rate generally accelerates during inspiration and slows down during expiration, creating respiratory sinus arrhythmia (RSA) which is the most prominent component of heart rate variability (HRV).[4] RSA serves a useful physiological function in improving respiratory gas exchange and circulatory efficiency, by matching ventilation and perfusion in the lung during each respiratory cycle.[5] Previous studies have shown that abnormal cardio-respiratory coordination, which is characterized by lower RSA, is associated with an increased risk of cardiovascular disease[67] and inefficiency of pulmonary gas exchange.[8]

Furthermore, since the vagal branch of the autonomic nervous system is involved in the coordinated responses from the respiratory and cardiovascular systems,[8] RSA is considered as an indirect indicator of cardiac vagal tone.[9] Botek and his colleagues have found that hypoxia reduces arterial oxygen saturation (SaO2) and vagal activity, with these decreases being pronounced in sensitive groups.[10] Slower oscillations in heart rate, low frequency (LF) HRV, reflect common action of the sympathetic and parasympathetic control centers.[11] The efficiency of vagal activity in modulating heart rate can be extracted from the short-term variation in RSA magnitude and heart rate.[12] In a study of wearable sensor data in patients with confirmed COVID-19, global measures of HRV that combine LF-HRV and RSA indicated a significant prediction of risk for hospitalization.[13] The specific method to estimate RSA magnitude has been shown to avoid many of the pitfalls of HRV research, including independence from respiratory parameters of rate and tidal volume, with respect to its estimation of cardiac vagal tone.[14] While the HRV parameters are known to be influenced by factors beyond acute disease exposure (e.g., age, weight), the protocol and metrics used optimize sensitivity for estimation of functional autonomic control of the heart. Therefore, RSA, LF, heart period (HP, inverse of heart rate), and the covariation of RSA and HP may be suitable noninvasive biomarkers to assess the changes of the autonomic nervous system in response to COVID-19, and individual differences in these parameters may be helpful in predicting the hospital LOS.

The present study aimed at investigating the possibility of using the demographic, clinical, and HRV parameters for forecasting the hospital LOS for COVID-19 patients. The associations between demographic, clinical, and HRV parameters were also assessed.

METHODS

Study design

All study aspects were reviewed according to the Strengthening the Reporting of Observational Studies in Epidemiology “STROBE” guideline. Consent was required and covered both study participation and publication of de-identified aggregate findings. No surrogate consent was allowed. De-identified individual subject data may be available from the corresponding author on reasonable request. We used a population-based prospective cohort study in which 41 patients with COVID-19 had their HRV measured on their 1st day of admission. They were followed until discharge and the correlation between HRV and other indices of disease with their hospital LOS were evaluated.

Setting

The study was conducted in a specialized hospital for respiratory disease, in which the COVID-19 patients were admitted. The 1st day of admission was considered to be marked by the patient having a clinical examination in the emergency department and being triaged. All patients were followed until relative recovery and discharged, as determined by clinical staff. The data collection was done in May and June 2020. Personnel who collected the data and investigators who analyzed the data were blinded to the result of the HRV-based measurements until after patients were discharged.

Participants

COVID-19 patients who were admitted during March and April 2020 to the Department of Infectious Diseases were included for this study. Because mechanical ventilation interferes with spontaneous breathing and impacts HRV,[1516] patients who were connected to the ventilator on the 1st day of admission were also excluded from the study.

Variables

On the 1st day of admission, study personnel collected electrocardiogram (ECG) recordings, arterial blood SaO2, and lymphocyte count (LC). These measurements were conducted after the patient's infection was confirmed by reverse transcriptase-polymerase chain reaction (RT-PCR) test. The RT-PCR tests were performed using commercial kits (BioGerm, Shanghai, China) according to the manufacturer's instructions.

At the time of admission SaO2, on room air without supplemental oxygen, was measured and recorded.

Data sources/measurement

ECG data were recorded in supine position for 7–10 min using a Polar H10 heart rate sensors (Polar Electro, Kempele, Finland).

Bias

We considered the hospital LOS and indicator for the severity of disease in patients who survived COVID-19. Other measures of COVID-19 severity including leukocyte count, SaO2, and CRP were also analyzed in the study. In addition, hospital bed management increases the importance of examining the hospital LOS in the specialized center for COVID-19 with resource limitations than other diseases severity factors. Charlson Comorbidity Index (CCI) was used to measure comorbidities and APACHE II scores were used to measure symptom severity.

Study size

A power calculation was conducted using the “pwr” package in R.[17] Given the novelty of the methods and application in this study, there were no prior published effect size estimates that could be used as a target effect size value. However, we calculated that given a sample size of 36 participants, the study would have 89% power to detect correlations of. 5 using two-sided tests at alpha = 0.05. Due to lack of prior data, power calculations for interactions with age could not be calculated prior to analysis, though values reported here may be used for power calculations in future studies.

Quantitative variables

Respiratory sinus arrhythmia amplitude

The vagus nerve is a major component of the parasympathetic branch of the autonomic nervous system.[12] RSA magnitude, a measure of cardiac vagal tone, is calculated from visually inspected interbeat interval (IBI) data. Artifact corrected IBIs are first evenly time sampled at five Hz. A fifty-one-point cubic Savitzky-Golay filter is applied to the five Hz time sampled data to estimate the slow periodic and aperiodic variance in the IBI time series. Fast-changing components of the signal are extracted from the difference between the filtered and source signals. Data are removed at the start and end of each recording to eliminate biases in the estimate of RSA magnitude. The difference signal is further constrained in the frequency domain to the frequency band of spontaneous respiration with an FIR bandpass filter (0.12–0.5 Hz). Discrete 20s-epochs are used to estimate the amplitude of RSA in ln (ms2). Average RSA value is calculated from these epoch values. This Porges-Bohrer method of RSA amplitude estimation has been shown in a study using partial vagal blockade to generate a stable estimate of cardiac vagal tone regardless of variations in respiratory parameters of rate and volume.[14]

Low frequency heart rate variability amplitude

LF-HRV reflects actions of both the sympathetic and parasympathetic nervous systems. Sympathetic modulation of vascular resistance around 0.06–0.10 Hz creates a pressure rhythm that is transduced by the baroreceptor reflex arc into changes in heart rate in this frequency band.[11]

A 131-point Savitzky-Golay filter is applied to the five Hz sampled IBI data from each subject to track the aperiodic variance below 0.04 Hz, which is then subtracted from the IBI time series. The output is bandpass filtered to attenuate RSA and yield a signal in the frequency range of 0.06–0.10 Hz. The mean of the natural log variance within discrete 20s epochs is used again to estimate the magnitude of the LF component of HRV for each subject.

Heart period

HP is calculated by taking the mean of IBIs over a period of 20 s epochs. HP is the reciprocal of heart rate and there is evidence that it has a stronger linear relation with autonomic control than heart rate and better distributional features for parametric analysis.[18] Epoch estimates of HP are extracted from the same segments of time as the RSA signal, accounting for data lost due to filtering, in order to facilitate calculation of vagal efficiency (VE).

Vagal efficiency

VE is an index of the dynamic influence of cardiac vagal tone on heart rate, calculated by the slope of the linear regression between simultaneous RSA and HP data drawn from each 20-s epoch.[19]

Statistical methods

Statistical analyses were conducted using R 4.0.1 (R Institute, Vienna, Austria).[20] Prior to analysis, variable distributions were examined for deviation from normality, and transformations were conducted as needed to conform to model assumptions. Bivariate associations were examined using Pearson correlation coefficients. Hospital days were modeled using Poisson regression. Robust standard errors were calculated using heteroscedasticity-consistent estimation of the covariance matrix.[21]

RESULTS

Forty-one participants were recruited for the study. Of these, one was transferred to another hospital and no hospital LOS data were available. Four participants had arrhythmias that did not allow for valid time series variability analysis. With the exclusion of these participants, 36 participants form the final study sample. Among these patients, 10 (27.7%) patients had underlying cardiovascular disease (hypertension, coronary heart disease, or cardiomyopathy), 3 (8.3%) patients had underlying respiratory disease (COPD, Asthma) and 10 (27.7%) patients smoked regularly. Of the 36 patients in this study, 38.9% had CCI zero, 16.6% had CCI one, and 44.4% had CCI ≥ two. The average APACHE II score in patients who were hospitalized fewer than 7 days (3.87 ± 0.63) was significantly lower than those hospitalized for more than 7 days (7.36 ± 0.87; p< 0.05). Three patients eventually required mechanical ventilation due to severity of disease, for on average duration of 8.3 (6–11) days. In addition, 6 patients required noninvasive ventilation for on average duration of 6.2 (4–9) days. Antibiotic therapy was administered for all patients. Twenty-nine (80.5%) patients were treated with Hydroxychloroquine and 20 (55.5%) patients received corticosteroids. Descriptive statistics are shown in Table 1.

Table 1

Descriptive statistics

Variable	n	Mean±SD	Median	Minimum	Maximum
HP (ms)	36	746.93±101.63	741.79	561.7	956.79
RSA	36	3.03±1.65	3.16	−0.15	7.76
LF	36	3.62±1.79	3.42	−0.65	6.68
VE	36	5.62±13.73	4.79	−27.46	31.44
LC	33	1484.07±713.23	1267	536.5	3553
CRP (ln)	30	1.99±0.76	2.1	0.69	4.56
Oxygen saturation	35	92.29±4.59	93	83	99
Hospital days	36	8.86±3.72	8.5	2	15

HP: Heart period, RSA: Respiratory sinus arrhythmia, LF: Low frequency, CRP: C-reactive protein, SD: Standard deviation, VE: Vagal efficiency, LC: Leukocyte count

Variable correlations are in Table 2. Older age was associated with lower RSA, lower LF-HRV, lower LC, lower SaO2, and longer hospital LOS. RSA and low vagal efficiency were related to higher leukocyte count. When age was included as a covariate, hospital LOS was no longer significantly predicted by leukocyte count (P = .306) and LF-HRV (P = .236).

Table 2

Correlations with 95% confidence intervals[1]

Variable	1	2	3	4	5	6	7	8
Age
HP	0.18 [−0.15-0.48]
RSA	−0.43** [−0.67-−0.12]	0.44** [0.13-0.67]
LF	−0.69** [−0.83-−0.46]	0.15 [−0.19-0.46]	0.74** [0.55-0.86]
5VE	−0.10 [−0.41-0.24]	−0.17 [−0.47-0.17]	0.13 [−0.21-0.44]	0.22 [−0.11-0.51]
LC	−0.53** [−0.74-−0.23]	0.08 [−0.27-0.41]	0.43* [0.10-0.67]	0.45** [0.13-0.69]	0.17 [−0.18-0.49]
CRP (ln)	0.20 [−0.18-0.52]	−0.20 [−0.52-0.18]	−0.34 [−0.62-0.03]	−0.36 [−0.64-0.00]	−0.18 [−0.51-0.19]	−0.23 [−0.55-0.16]
Oxygen saturation	−0.47** [−0.69-−0.16]	−0.19 [−0.49-0.16]	0.23 [−0.11-0.53]	0.27 [−0.07-0.55]	0.14 [−0.20-0.46]	0.24 [−0.12-0.55]	−0.37* [−0.65-−0.00]
9Hospital days	0.59** [0.33-0.77]	0.30 [−0.03-0.57]	−0.27 [−0.55-0.07]	−0.53** [−0.73-−0.24]	−0.30 [−0.57-0.03]	−0.48** [−0.71-−0.17]	0.34 [−0.02-0.62]	−0.59** [−0.77-−0.32]

*P<0.05, **P<0.01. Values in square brackets indicate the 95% CI for each correlation. CI: Confidence interval, HP: Heart period, RSA: Respiratory sinus arrhythmia, LF: Low frequency, VE: Vagal efficiency, LC: Lymphocyte count, CRP: C-reactive protein

Modeling by age group

Given the strong association between age and hospital LOS, modeling tested the interaction of predictor variables with age group to predict hospital LOS. Modeling was conducted using Poisson regression with robust standard errors. The age reference category is the youngest age group (<40 years). Standardized estimates were calculated with the predictors on a standardized metric (M = 0, standard deviation = 1). Standardized percent change differences were calculated using the exponential transformation of the standardized estimate (e^x for positive values and 1-e^x for negative values).

There were no interactions between age category and RSA, HP, LF-HRV, or leukocyte count, indicating limited utility for these metrics in predicting hospital LOS (days). However, there were differences in the prediction of hospital LOS by vagal efficiency as shown in Figure 1. A goodness of fit test, calculated from the residual deviance, supported the model as fitting the data well (residual deviance [SD] =22.27, df = 29, P = .809). Higher vagal efficiency was associated with a shorter hospital LOS in the youngest age group only [Table 3; P = 0.007] wherein a one SD higher vagal efficiency was associated with a 19.27% shorter hospital LOS. There was no association in the middle age group (P = 0.547) or the oldest age group (P = 0.0525).

Figure 1

Predicted values and 95% confidence interval of hospital days predicted from vagal efficiency by age group

Table 3

Count of hospital days predicted by the interaction of vagal efficiency and age category

Term	Estimate	Robust SE	STD estimate	STD % difference	P
Model 1
Intercept	1.904	0.093	1.8165		<0.0001
Vagal Efficiency Slope	−0.016	0.006	−0.2137	−19.27	0.007
Age category: 40-60 years	0.291	0.223	0.3346	33.78	0.191
Age category: >60 years	0.584	0.109	0.6613	79.32	<0.0001
Interaction terms
VE slope × age category: 40-60 years	0.008	0.014	0.1054		0.592
VE slope × age category: >60 years	0.014	0.006	0.1882		0.034
Model 2
Intercept	11.0746	2.2726	2.0657		<0.00001
Oxygen saturation	−0.0976	0.0245	−0.4477	36.09	<.0001
Age category: 40-60 years	−8.6191	2.4645	0.0077	0.77	0.0005
Age category: >60 years	−3.5206	2.6574	0.4200	52.18	0.1852
Interaction terms
Oxygen saturation× age category: 40-60 years	0.0382	0.0288	0.1753		0.1846
Oxygen saturation × age category: >60 years	0.0979	0.0267	0.4492		0.0002

STD: Standardized, SE: Standard error, VE: Vagal efficiency

In addition, there were age group differences in the prediction of hospital LOS from SaO2 [Table 3]. Model predictions and raw values for each age group are shown in Figure 2, computed with each age group set as the reference category. High SaO2 predicted shorter hospital LOS in the youngest age group (<40 years) with one standard deviation higher SaO2 associated with a 36.09% shorter hospital LOS (P < 0.001) and the middle age group (40–60 years) in which a one SD higher value was associated with a 23.81% shorter hospital LOS (P < .001). There was no significant prediction in the oldest age group (>60 years, P = 0.975).

Figure 2

Predicted values and 95% confidence interval of hospital days predicted from SaO2 by age group

DISCUSSION

In this study, the demographic and clinical evaluations showed that there was a significant correlation between the hospital LOS and four variables: LC, age, SaO2, and LF-HRV. First, there was a negative correlation between the hospital LOS (days) and LC, since LC was lower in patients with more severe COVID-19 disease and longer hospital LOS. Similar to Liu et al., we explored demographic factors, laboratory findings, and treatment data to predict the hospital LOS in COVID-19 patients.[2] According to their study, of all the parameters examined, only the reduction of LC was introduced as a risk factor for long-term hospital LOS of patients. Past studies have suggested three reasons for reducing the number of lymphocytes in COVID-19 patients. (1) The virus may directly infect lymphocytes and cause the death of these cells.[22] (2) The virus may directly attack lymphatic organs, such as the thymus and spleen. (3) Releasing large amounts of inflammatory cytokines may lead to apoptosis of lymphocytes.[23] However, some studies have shown that LC is reduced in only 65% of COVID-19 patients.[24] Therefore, other parameters are needed to predict the hospital LOS of COVID-19 patients.

Second, the present study demonstrated that the older patients have a prolonged hospital LOS. Similar to our results, Qu et al. have reported a positive relationship between COVID-19 patients' age and the hospital LOS, and a negative relationship between LC and the hospital LOS of COVID-19 patients.[25] However, according to the study of Liu and his colleagues, no significant relationship was found between patients' age and the hospital LOS.[2] In addition to sample limitations in terms of size and geographic diversity, this contradictory result may be due to differences in the number of elderly and critically ill patients in each study population, or this may be due to higher mortality of older patients.

Third, according to our results, there is a negative relationship between SaO2 and the hospital LOS, but no study was found in literature that has mentioned any relationship between SaO2 and the hospital LOS of COVID-19 patients.

Fourth, HRV is a noninvasive technique used to evaluate autonomic disorders and other pathological conditions such as metabolic syndrome, diabetes,[26] and acute respiratory distress syndrome.,[27] In addition, previous studies have shown that HRV changes in response to infection, and the degree of these changes may be proportional to the severity of the infection and may predict poor outcome of infection.[28] Therefore, it may be possible to predict the hospital LOS of COVID-19 patients with HRV analysis.

Limitations

Our study had the following limitations. First, the recording of electrocardiography was very difficult in this study, due to strict infection control precautions in this patient population and relative shortage of personal protective equipment, resulting in a small sample. Despite the small number of patients, we observed a significant difference in some demographic, clinical, and HRV parameters between patients who were hospitalized for a short time and patients who were hospitalized for a longer period of time. Second, it is observed that air temperature, noise, and light might affect the heart rate, respiratory rate, and HRV parameters.[293031] Therefore, patients' ECG was recorded non-invasively, in full consciousness, in a supine position, and away from environmental stress (surrounding temperature, light, and noise) to the extent possible. Additional studies, limited to ICU patients but including those with fatal cases, may be needed to fully characterize the relationship between HRV based predictions and disease severity. Our sample size was also small and larger studies may be needed.

Our analysis revealed that there was a negative relationship between the LF-HRV component and the hospital LOS of COVID-19 patients. In fact, the LF-HRV component was lower in critically ill patients who need longer-term hospital care. The loss of LF-HRV amplitude, during the stability of the parasympathetic indices, may be due to decrease in vasomotor activity or autonomic (especially sympathetic) regulatory impairment.[32] Our results may support a previous study reported that the pattern changes in LF component of blood pressure and heart rate signals can predict poor outcome in critically ill patients.[33]

While there was no overall relationship between the parasympathetic indicators of heart rate (RSA and VE) and the hospital LOS of COVID-19 patients, after dividing the patients into different age groups (<40, 40–60 and >60), it was observed that in the younger age group, the hospital LOS of patients decreases with increasing vagal efficiency. Higher vagal efficiency reflects a tight link between variations in cardiac vagal tone and heart rate and implies a cardio-inhibitory system that is not being counteracted by other mechanisms that would lower efficiency, such as sympathetic activation.[5] The mechanism underlying this relationship in critically ill patients has not been determined. However, previous studies have shown that electrical stimulation of the vagus nerve during fatal endotoxemia can inhibit tumor necrosis factor α synthesis and prevent the exacerbation of inflammation.[34] Therefore, increased dependence on vagal modulation in young COVID-19 patients may be their pathophysiological response to inhibit cytokine synthesis, modulate systemic inflammation, and prevent organ dysfunction; which ultimately reduces the hospital LOS in these patients.

Based on our results in middle and older age groups, differences in vagal efficiency have no significant effects on the hospital LOS. Some studies have shown that with increasing age, the responses of nicotine receptors (especially alpha7nAChR)[35] show a gradual decline with time. Since these receptors are involved in the cholinergic anti-inflammatory pathway, their dysfunction may diminish the role of the vagus nerve in preventing excessive inflammation. Thus, there may not be a hospital LOS benefit in middle-aged and elderly patients who have higher vagal efficiency.

The sample in this study was recruited in a hospital setting with adult patients from the community who were receiving care for COVID-19, which suggests strong generalizability for the environment in which the results would be applicable. However, additional studies with larger samples are needed to demonstrate that the finding can be replicated.

CONCLUSIONS

It may be possible to noninvasively assess HRV and vagal efficiency in younger COVID-19 patients to tailor treatment options for those who show indicators of rapid or prolonged hospital LOS. Similar age dependence was found for SaO2 as a prognostic indicator, with maximal utility for younger patients.

Understanding and predicting hospital bed demand is essential for decision-making and treatment planning in pandemic situations such as the COVID-19 pandemic. In most countries, the rate of hospitalized patients with COVID-19 is estimated by epidemic curves. However, because COVID-19 has various levels of presentation and severity, the hospital LOS may be strongly influenced by patient characteristics. Therefore, the aim of this study was to investigate the possibility of using clinical manifestations, laboratory data, and HRV parameters of COVID-19 patients to predict hospital LOS of patients. Our results have shown that it may be possible to noninvasively assess HRV and vagal efficiency in younger COVID-19 patients to inform treatment options for those who show indicators of rapid or prolonged hospital LOS. Similar age dependence was found for SaO2 as a prognostic indicator, with maximal utility for younger patients.

Research quality and ethics statement

This study was approved by the Institutional Review Board / Ethics Committee at Shiraz University of Medical Sciences, (IRB/Ethics approval number is IR.SUMS.REC.1399.95; Approval date April 22, 2020). The authors followed the applicable EQUATOR Network (http://www.equator-network.org/) guidelines, specifically the STROBE Guidelines, during the conduct of this research project.

Financial support and sponsorship

Research Council of Shiraz University of Medical Sciences (grant No 99-01-106-22233).

Conflicts of interest

There are no conflicts of interest.