Trajectories of Key Physiological Parameters in COVID-19 Patients Using Continuous Remote Monitoring and Health AI

Coronavirus disease 2019 (COVID-19) exerts deleterious effects on the cardiorespiratory system, leading to worse prognosis in the most effected. The aim of this retrospective multi-center study was to describe the variability of key cardiopulmonary vitals amongst hospitalized COVID-19 patients, measured every 15 minutes using a novel wearable chest-monitor. A total of 492 patients were included, with >3 million measurements collected including heart rate, systolic and diastolic blood pressure, cardiac output, cardiac index, systemic vascular resistance, respiratory rate, blood oxygen saturation, and body temperature. We show differential trajectories of these vital signs, apparent within the rst 24hrs of monitoring. Importantly, we show for the rst time that cardiovascular deterioration appears early after admission and in parallel with changes in the respiratory parameters, and identify sub-populations at high risk. Combining frequent monitoring using wearable technology with advanced big data and AI analysis tools may aid early detection of deterioration of COVID-19 patients. biomarkers and echocardiography directed at cardiac injury. In this study, and for the rst time, we used continuous and long-term monitoring of advanced cardiac parameters using a novel and non-invasive technology, previously showing its capabilities with similar measurements compared to invasive techniques 44 . We found several interesting components to the cardiovascular changes, such as a decrease in CI among the risk populations (e.g. male, elderly and obese patients) after ve days of monitoring, dropping to lower than the normal range in males, in addition to an increase in SVR among these populations. Previously published data showed direct damage to the heart and its function, as well as thrombogenesis which might inuence the SVR. However, we do not have sucient data on these parameters, in COVID-19 patients or on the general population, and this should be studied further.


Introduction
Several months after declaring coronavirus disease 2019 (COVID-19) a global pandemic, millions became infected all over the world, with over a million deaths due to the virus and numbers still rising. Covid-19 is a multi-system disease with a wide range of clinical manifestations, from asymptomatic patients, through simple in uenza-like illness, and all the way to a fulminant disease comprised of severe respiratory involvement with acute respiratory distress syndrome (ARDS) and pulmonary insu ciency 1 . As the understanding of the disease evolved, it was shown that COVID-19 damages epithelial and endothelial cells in numerous tissues leading to SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2)-related multi-organ failure [1][2][3][4][5] . Additional damages caused by SARS-CoV-2 may include neurological diseases and hypercoagulable state [6][7][8] .
Under the burden of this pandemic, resources are limited and the fear of infection transmission is high. Novel technologies may expand the capacity to remotely monitor patients, providing optimized support and protection both for COVID-19 patients and for health care providers by reducing direct contact without compromising the treatment given to isolated patients, as well as lightening the strain on the medical teams and providing early detection of patient deterioration 25,26 . Continuous monitoring of multiple vital physiological parameters may allow better care for patients who are acutely ill, for hospital-at-home of stable patients, and for early discharge of admitted patients 25,27 . It is also accepted that such a system would be of bene t if it frequently and automatically measures numerous vital signs, improves patient surveillance and, ultimately, improves patient outcomes 26,[28][29][30] . While standard vital sign measurement is executed over a single short time period and might miss crucial changes in physiological parameters, frequent remote patient monitoring (RPM) systems are potentially better equipped to detect and alert of changes since the vital signs measurements are taken continuously and for longer periods 26,31,32 .
Several reports so far have identi ed differences in COVID-19 infection rates, symptom severity and mortality between sex, age, and body mass index (BMI) categories [33][34][35][36][37][38][39][40] ; however, whether the physiological response of key cardiovascular and respiratory parameters during the course of hospitalization is different between individuals remains unknown. This knowledge may help in further understanding the clinical course of the disease and preventing clinical deterioration, with highlighted importance due to the rapid deterioration often seen in COVID-19 patients 1 .
The aim of this retrospective multi-center study was to determine the trajectory of nine key and advanced physiological parameters amongst COVID-19 patients admitted to isolation units in Israeli medical centers. These parameters were continuously collected using a wearable, non-invasive, wireless chest-monitor (Biobeat Technologies Ltd., Petah Tikva, Israel; Figure 1), and big data analysis was conducted using advanced AI and bioinformatic tools. A special emphasis was given to characterizing the disease progression among these patients, and the identi cation of differences in physiological responses over time of sub-groups according to age, sex, and BMI.

Results
Initially, 571 patients participated in the study. Subjects with less than 24hrs of continuous tracking data were excluded, with 492 patients remaining in the nal analysis. As can be seen in Figure 2 a-c, the number of patients decreased throughout the monitoring period due to patients' discharge from the hospital, transfer to intensive care units (ICU), or death. As a result, we have decided to focus our analysis on the rst ve days (120hrs) after admission. Physiological measurements were recorded continuously every 15 minutes with a total of 3,215,334 measurements. For the raw, un ltered data, measurements were collected during an average monitoring period of 75.26hrs (range 0-455), with 245.67±226.39 observations (±standard deviation) per patient. The analyzed ltered data included patients with at least 24hrs of tracking, and included observations from the rst ve days only. Measurements were collected during an average monitoring period of 74.78hrs (range 24-120) per patient (174.89±110.58).

Analysis of baseline parameters
Subject characteristics, along with mean values during the rst 2hrs of monitoring (average of 8 measurements) for body temperature, SpO 2 , RR, heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), cardiac output (CO), cardiac index (CI), and systemic vascular resistance (SVR) for each sub-group are presented in Table 1. The rst 2hrs served as a baseline monitoring period for each subject. Men (64% of the recruited patients) had a signi cantly higher BMI and weighed more than women (p < 0.001). Middle aged patients (men = 70%) had signi cantly higher weight (p < 0.001) than young (men = 64%), elderly (men = 66%) and >80-year-old (men = 44%) patients, and no differences in weight were detected between the three other age groups. Young patients had signi cantly (p < 0.01) lower BMI compared to the other groups, and no differences were noted between middle age, elderly, and >80-year-old patients. Patients with normal BMI (men = 56%) were signi cantly (p < 0.01) younger.
For baseline physiological measurements, an ANOVA analysis revealed a signi cant (p < 0.01) difference between men and women for body temperature, SpO 2 , HR, DBP and CO. When analyzing age groups, signi cant (p < 0.003) interactions were seen for all physiological measures recorded during the rst 2hrs of admission. Lastly, only SpO 2 , RR, CO and CI were signi cantly (p < 0.01) different between BMI groups upon admission.
Trajectory of physiological vitals over ve days from admission Figure 3 provides an overall description of the nine vital signs in the 130 patients that fully completed ve days of continuous monitoring. Overall, within the rst 24hrs we found a signi cant increase in temperature, RR, and SVR, and a signi cant decrease in SpO 2 , DBP, CO and CI (p < 0.01 for all). These changes all appeared at the same time. For HR, SBP, CO and CI, the changes appeared in a repetitive pattern.
Further analysis is provided in Figure 4, where we show the results of repeated measures ANOVA tests performed to determine differences between groups during the same timeframe.
Body temperature (Figure 4a): In both males and females, temperature increased during the rst 24hrs of monitoring, signi cantly higher among males (p < 0.001). From the second day, females showed a signi cant decrease in temperature until the fth day (p < 0.001), while males' trajectory of temperature remained similar. Across the BMI groups, temperature increased similarly during the rst 48hrs. A slight increase was noted in the normal weight group and decrease among the overweight and obese, but by the fth day there were no differences within the BMI group. In the age group, temperature among the elderly was higher (p < 0.001) during the whole 5 days. SpO 2 (Figure 4b): SpO 2 decreased in both males and females during the rst 48hrs (p < 0.001 for both), with a higher decrease among males. Starting from 72hrs since admission, and throughout the next two days, females showed a quicker return to the baseline levels, while males remained with lower values (p < 0.03 between sex). Among the BMI sub-groups, similar trajectories were found, with higher SpO 2 levels in the normal weight sub-group, a lesser level in the overweight, and the lowest SpO 2 levels among the obese. In the age group, the young maintained SpO 2 levels throughout the ve days of hospitalization with a slight reduction (p = 0.044), while other sub-groups showed dynamic changes with lower SpO 2 values.
Respiratory rate (Figure 4c): Within the rst 48hrs from admission, RR increased signi cantly in males, obese patients, and in all age subgroups (p < 0.01) except the young. At ve days after admission, there were no signi cant differences between sex. During the whole ve days of hospitalization, the overweight and obese sub-groups had higher RR as compared to the normal weight sub-group (p < 0.01). The young showed a signi cant increase (p = 0.044) in RR, and with no signi cant changes among the other sub-groups.
Heart rate ( Figure 4d): During the rst 48hrs, HR dropped among men, while increased among women (p < 0.01 and p < 0.01, respectively). From then on, the dynamics of HR behaved in opposite manners between sex, until day 5, in which both showed an increase, with males not reaching the baseline values and females showing a signi cant increase from baseline (p < 0.01). Young patients showed a mild increase during the rst 24hrs, remaining stable until day 5, during which the levels decreased back to baseline levels. Middle aged and elderly had a gradual decrease until day 4, and both increased to levels similar to those at baseline at day ve. Patients over 80 years had peak increases at 24 and 48hrs after admission, followed by a sharp decrease to baseline levels in day 3, followed by an increase until the end of day 5. Though patients in the age groups started with the same baseline, after 24hrs the normal weight and obese started to show dynamic decrease until the 5 th day, while the overweight had an overall increase.
Systolic Blood Pressure (Figure 4e): SBP showed similar trends among sex, with higher levels among males at day 4 after admission, returning to similar values by day 5. Changes in SBP over ve days of hospitalization were signi cantly different between age groups (p = 0.002). The young showed a decrease in SBP during the rst 24hrs, remaining relatively stable during the next three days, followed by an increase in SBP to a level higher than the baseline on day 5. The middle age and the elderly were relatively stable during the ve days of monitoring. The over 80 years sub-group had higher SBP values during the whole period, increasing to peak levels by 48hrs, showing a relatively sharp decrease to baseline levels on day 4, and increasing back on day 5. Obese patients showed a decrease trend in SBP until day 4, increasing back to baseline levels on day 5. Normal weight and overweight patients showed a changing pattern with a decrease trend in SBP, both reaching nadir on day 4, and showing increase to baseline levels on day 5.
Diastolic Blood Pressure (Figure 4f): Both sexes showed a decrease in DBP values during the rst 4 days, women to a higher extent than men (p < 0.001), and in the fth day both showed increase with values among women returning to the baseline level. Patients over 80 years showed relatively lower DBP values during the 5 days period, reaching a nadir at day 4 (p < 0.001). The young and elderly showed a decrease trend in the rst 3 days (p < 0.01), returning close to baseline levels by day 5. Though the middle age sub-group had the highest DBP levels, they also had a similar trend of decrease until day 4, returning close to baseline level on day 5. Normal weight patients had a sharp decrease during the rst 24hrs (p < 0.001), maintaining this level until the 4 th day, followed by a sharp increase during the 5 th day to baseline levels. Both overweight and obese had a milder decrease during the rst 4 days with overweight returning to baseline levels in the 5 th day, and the obese returning to a lower level than the baseline (p = 0.02).
Cardiac Output (Figure 4g): Males had a sharp decrease in the rst 24hrs (p < 0.01), kept stable in the next 48hrs, followed by a sharp decrease in the 4 th day, and an increase in the 5 th day to a level below baseline. Females showed dynamic changes reaching peak high levels on each of days one to three, followed by a drop shortly after every peak, followed by a constant increase until reaching the highest peak on day 5 (p < 0.01). Young patients showed stable CO values during the whole ve-day period. Middle age and elderly showed a slight decrease during the rst 24hrs, followed by a dynamic trend, with middle aged returning to the baseline levels while the elderly did not show signi cant changes from baseline (p = 0.081). Patients over 80 years had a dynamic pattern of changes with peaks at 24hrs and 48hrs from admission (p < 0.01 and p = 0.02, respectively), followed by lower peaks and increasing again at day 5 after admission. All BMI sub-groups had decreased levels of CO in the rst 24hrs.
Cardiac Index (Figure 4h): Both sexes started with the same CI values at baseline. Shortly after, a sharp increase was evident amongst females reaching its peak at 48hrs after admission (p < 0.001), while amongst males a sharp decrease was seen after 24hrs (p < 0.001), maintained until day 4, in which a further decrease was evident, and a moderate increase appeared on day 5, to levels lower than baseline (p < 0.001). Amongst the young, dynamic changes were not statistically signi cant, with similar values to baseline measured on day 5.
Both middle aged and elderly showed decreases within the rst 24hrs (p < 0.001 in both). From that moment on, the middle aged showed moderate increase, becoming more pronounced on the 5 th day, returning to baseline levels. However, amongst the elderly, further decrease was evident on day 4, increasing slightly on day 5 (p < 0.001). Unlike other age sub-groups, the over 80 years started with a sharp increase in CI during the rst 24hrs, followed by a sharp decrease and immediately followed with a higher increase by 48hrs after admission (p < 0.001). This was followed by an unstable decrease over the next two days, and a sharp increase at day 5 (p < 0.001). Normal weight and overweight patients had dynamic non-signi cant changes in CI, with values similar to the baseline at the end of day 5. However, obese had a lower baseline level of CI, further decreasing during the rst 24hrs, and reaching a nadir at day 5 (p < 0.001).
Systemic Vascular Resistance (Figure 4i): Females had a higher SVR value at baseline as compared with males, and during the 5-day period showed consistent decrease until the end of monitoring (p < 0.001). Males started with an increase during the rst 24hrs (p < 0.001), and remained relatively stable until day 5. Over 80 years started at a higher level of SVR at baseline, and showed continuous decrease until day 5. Other age sub-groups started at the same baseline, with the middle age showing an increase in the rst 24hrs (p < 0.01) followed by continuous decrease until day 5, returning to baseline level. The elderly remained stable until an increase on day 4, returning to baseline levels on day 5. The young showed dynamic decrease until a nadir on day 4, followed by an increase on day 5 back to baseline levels. Normal and obese patients had dynamic changes over the ve-day period without signi cance. Overweight patients showed continuous decrease during the ve-day period (p < 0.01).

Discussion
Frequently measured vital signs might help improve the understanding of the progression of COVID-19 in humans. So far, there were no reports detailing frequent monitoring of vitals in COVID-19 patients for prolonged periods of time, nor reports including advanced vitals such as CO, CI, and SVR, neither are there any reports that analyze how the disease in uences vital sign trends. In Israel, many COVID-19 isolation units used (and are still using) a novel PPG-based platform for RPM (Figure 1). In the current multi-center study, we present the results of analyzing nine physiological parameters (temperature, SpO 2 , RR, HR, SBP, DBP, CO, CI, and SVR) continuously monitored for ve days in 130 COVID-19 patients within isolation units in ve medical centers. This allowed an in-depth analysis and a more comprehensive understanding of the disease progression, focusing on sub-groups in danger of higher morbidity and mortality. It is established that advanced age, male sex, and overweight are worse prognostic factors in COVID-19 [33][34][35][36][37][38][39][40] .
In this study, we show that this decline manifests upon admission with inferior cardiorespiratory parameters, and during the hospitalization period with increased average temperature in males, lower SpO 2 , and increased RR in males, elderly patients and overweight patients.
We found that during the rst 24hrs from admission the changes in the cardiovascular parameters appear in parallel to the changes in the respiratory parameters, an insight that was not reported previously. Moreover, the decrease found in DBP throughout most of the monitoring period might correlate with prior reports on diastolic dysfunction resulting from COVID-19 damage to the heart 21,22 , yet this hypothesis is still to be validated since we did not perform echocardiography. This could also be regarded as clinical deterioration, presented with concomitant decrease in CO and CI, and an increase in RR. The changes we found were more prominent amongst males, patients older than 80 years, and obese patients. This could also explain why patients with preexisting respiratory and cardiovascular diseases have an increased risk of severe morbidity and mortality 1,6,[11][12][13][14]18,19,33,34,39,[41][42][43] .
Previous studies employed periodic and infrequent biomarkers and echocardiography directed at cardiac injury. In this study, and for the rst time, we used continuous and long-term monitoring of advanced cardiac parameters using a novel and non-invasive technology, previously showing its capabilities with similar measurements compared to invasive techniques 44 . We found several interesting components to the cardiovascular changes, such as a decrease in CI among the risk populations (e.g. male, elderly and obese patients) after ve days of monitoring, dropping to lower than the normal range in males, in addition to an increase in SVR among these populations. Previously published data showed direct damage to the heart and its function, as well as thrombogenesis which might in uence the SVR. However, we do not have su cient data on these parameters, in COVID-19 patients or on the general population, and this should be studied further.
Another behavior of note is that the trends seen in HR, SBP, CO, and CI seem to be repetitive (Figure 3 and 4). This might be part of a circadian rhythm, yet again -we do not have enough data to substantiate this observation, and it should be further studied.
Signi cant changes between the studied groups were already apparent during the 2-hour baseline period, emphasizing that physiological effects of COVID-19 are different among the various groups beyond the expected naturally-occurring differences.
Although statistically signi cant changes were found in some vitals among the groups, they were slight and are currently not considered to be of clinical signi cance. However, we think that as advanced monitoring tools will keep developing and being introduced into clinical practice, ampli ed by advanced AI and machine learning analysis tools, we may nd that even these slight changes could have signi cance and warrant clinical attention, especially when developing early warning score systems in the context of complex patients.
Operationally, the continuously collected data was transferred automatically and in real-time every 15 minutes to the medical staff. This reduced the direct contact between medical staff and patients without compromising the medical care provided, an important feature highly required during a pandemic.
When looking at the ambulatory and out-of-hospital environments, the early identi cation of symptomatic and pre-symptomatic infected individuals would be especially valuable in breaking chains of infections, principally due to increased transmission during this period [44][45][46][47] . A small wearable, wireless RPM device that frequently collects and transmits the data automatically and in real-time could help in achieving that. Moreover, by using this technology with COVID-19 patients we now have an opportunity to de ne a novel COVID-19 score for the accurate early detection of deterioration. Due to sex-, age-, and BMI-related differences in disease progression [33][34][35][36][37][38][39][40] , it might be prudent to de ne a separate score for individuals based on their basic characteristics. This tailored precision medicine approach should be further studied.
A limitation of this study is that we did not have the clinical data regarding the course of hospitalization and nal outcomes of these patients. All were admitted to isolation units in moderate to severe condition, and some were later transferred to COVID-19 dedicated ICUs.
We have no information regarding outcomes, administration of supplemental oxygen, vasopressors, and speci c therapeutics against COVID-19. However, they all received advanced medical care, and the number of data points of multiple physiological parameters collected was large and frequent, still allowing to have insights of clinical signi cance. On-going studies are now conducted in order to allow the parallel analysis of collected vitals and clinical data.
In conclusion, in this study we have shown for the rst time the trajectories of cardiorespiratory parameters during a long period of frequent monitoring using an RPM system. Using advanced bioinformatic tools we demonstrate that populations at risk display worse respiratory parameters, as well as worse advanced cardiovascular parameters, emphasizing the cardiorespiratory effects of COVID-19 over time, with differential physiological responses noted between sex, BMI and age groups. This may serve to improve early detection of clinical deterioration of COVID-19 patients, especially important in times of overwhelmed health care systems, helping to reduce direct contact between health care providers and COVID-19 patients without compromising medical care.

Study Design and Overview
This retrospective non-interventional study was conducted between March 3 rd , 2020 and May 22 nd , 2020. 571 COVID-19 patients were

PPG-based Device
The chest-monitor device used in this study ( Figure 1) utilizes a unique re ective PPG technology, in which speci c wavelengths of light are transmitted onto skin tissue, re ected from the tissue and detected by a photodiode detector positioned near the light source transmitter. The sensor tracks vital signs derived from changes in the pulse contour, following a simple offset baseline trimonthly calibration process using an approved non-invasive, cuff-based device, and is based on Pulse Wave Transit Time (PWTT) technology combined with Pulse Wave Analysis (PWA) 25,48,49 .
Patients were monitored throughout hospitalization time, and patches were replaced if hospitalization lasted longer than the 6-day battery life of the patch. Same patches were not used to monitor different patients.

Statistical analysis
Global outliers were picked using PCA and histogram examination. On a per-patient level, outlier observations with difference from the mean that's greater than Q3 + 3xIQR, were discarded.
All statistical analyses were made using R version 3.6.3 50 . Differences between the studied groups were determined using independent ttest when the data satis ed test requirements. Equal variances were assumed if Barlett and Levene tests came out signi cant. Wilcox test was performed for metrics with non-normal distribution. For key physiological variables recorded during the rst 2hrs of admission, normality was assessed using the Shapiro test QQ-plots, allowing the removal of extreme outliers. Then, differences between groups were determined using repeated measures ANOVA. Where a signi cant main effect was found, a Tukey's post hoc test was performed. The data was tted to a linear mixed model with nlme 3.1, with sex, age range and BMI as coe cients, then tested using ANOVA. Pairwise Wilcox was used for post-hoc testing when the sample-sizes were su ciently large (n>50). Trend estimation gures with less than 1,000 observations were done using loess, while larger trend data was done using gam. All other results are presented as means ± SD. Statistical analyses were considered signi cant if p < 0.05.    Average recorded measurements among 130 patients continuously monitored for 5 days from admission, without separation to subgroups. Temp, temperature; SpO2, blood oxygen saturation; RR, respiratory rate; HR, heart rate, SBP, systolic blood pressure; DBP, diastolic blood pressure; CO, cardiac output; CI, cardiac index; SVR, systemic vascular resistance. The blue line represents mean value of each vital and the 95% con dence interval appears in gray.

Figure 4
Measured vital signs by Sex, BMI and Age group, over the course of the rst 5 days of tracking. De nitions of BMI and Age groups are provided in Figure 2. Temp, temperature; SpO2, blood oxygen saturation; RR, respiratory rate; HR, heart rate, SBP, systolic blood pressure; DBP, diastolic blood pressure; CO, cardiac output; CI, cardiac index; SVR, systemic vascular resistance. Each line represents mean values of the vitals and 95% con dence interval appears in gray.