Among patients in the United Kingdom hospitalized with COVID-19 during the pandemic, 42% required ventilatory support or high-dependency or critical care or died in hospital [1]. Therefore, patients diagnosed with COVID-19 managed in hospital wards required close monitoring of vital signs to ensure that deterioration was identified promptly and care was escalated appropriately. In usual care, patients at high risk of deterioration are placed in high-visibility areas of the ward so that nurses can observe them closely. However, due to the elevated risk of COVID-19 transmission both to other patients and staff, patients were often cared for in single isolation rooms or closed cohort bays [2]. Isolation, plus the time required by staff to don personal protective equipment (PPE), posed challenges to the close observation required to ensure the safety of these patients who are acutely ill.

Continuous monitoring of vital signs using wearable devices has long been recognized as having the potential to detect deterioration earlier than intermittent measurements [3-5]. Early in the pandemic, it was recognized that patients diagnosed with COVID-19 had elevated respiratory rates and rapidly worsening oxygen saturation levels that required high levels of supplemental oxygen [6]. As a result, the use of wearable vital sign monitoring escalated, with devices commonly used to allow low-risk patients to be managed at home [7].

At the start of the COVID-19 pandemic, we were close to completing development of a wearable system for the continuous monitoring of heart rate, respiratory rate, and oxygen saturation in hospital wards as part of a research project. Multiple phases of development had been completed that assessed the wearability, accuracy, and reliability of the wearable devices and system [8-10]. Given the concerns about maintaining patient and staff safety in wards with patients with COVID-19, it was clear that this system had the potential to support local clinical care in wards receiving patients with COVID-19. Implementation was therefore accelerated, and the system was deployed in a ward in late March 2020, the day on which the United Kingdom first entered lockdown, with the aim of supporting nurses in managing these high-risk patients.

Although wearable monitoring is increasingly being used in ward-based care internationally to support patient safety [3], little is known about the mechanism of impact [5]. Studies to date have focused on improving patient safety–related outcomes (such as in-hospital mortality and length of stay) [3], whereas studies of staff and patient views of wearable monitoring system (WMS) use have focused on safety and reassurance [11-13]. The presence of WMSs alone cannot improve patient outcomes, and the impact of WMSs on nurse behaviors related to vital sign monitoring is not well understood [5].

The primary aim of this retrospective data study was to describe the physiological pattern of vital signs over the course of SARS-CoV-2 infection for hospitalized patients outside of critical care. In this paper, we also report a post hoc analysis exploring the impact of displaying data from the WMS on the timing of nurses recording intermittent manual vital sign measurements (MVSMs).

The service evaluation was undertaken in an isolation ward with 19 side rooms and two 24-bed medical wards where patients with COVID-19 were cohorted in open bays within a large tertiary referral hospital in the United Kingdom.

We included all patients diagnosed with COVID-19 (confirmed via laboratory polymerase chain reaction test) and remotely monitored at any point during their stay in 1 of the 3 wards where we deployed the wearable system between March 23, 2020, and February 28, 2021.

Pseudonymized data were retrospectively extracted from the local National Health Service (NHS) hospital data warehouse by hospital employees and associated with the wearable data via a link table (mapping local record numbers to system identifiers) after applying the NHS digital data opt-out service. All data were stored on a secure hospital server for analysis and only accessible within the hospital network.

Data extraction was limited to the following variables:

Extracted data were limited to the duration of hospital admission. There was no follow-up, and study duration ended at death or hospital discharge.

All patients had periods without continuous monitoring data before and/or after being on the WMS. These data were split into 2 datasets: periods A (periods without wearable monitoring data—“unmonitored periods”) and B (periods during which wearable monitoring data were recorded—“monitored periods”). Therefore, the 2 datasets were generated from the same cohort of patients. Given the retrospective design, it was not possible to access intermittent vital sign data from a nonmonitored cohort of patients.

Patients with no monitoring data were not included in the analysis. Once a patient was considered to have been monitored using wearable devices, data availability was calculated as the percentage of at least one vital sign value being received within a 5-minute interval between the first and last wearable data point received.

As this was a retrospective data study that did not use any identifiable data, Health Research Authority approval was sought and granted (12/HRA/1234), and the study was registered (ISRCTN 85624923).

A total of 157 unique patients were monitored using the WMS between March 23, 2020, and February 28, 2021, with data on 144 (91.7%) patients available after exclusions were applied (Figure 2). There were no missing outcome data from included patients. The median age was 62.5 (IQR 47-77) years, and 29.9% (43/144) of the patients were female. The age distributions of patients admitted during waves 1 and 2 are shown in Figure S5 in Multimedia Appendix 1. Of the 144 patients included in the analysis, 97 (67.4%) were admitted to side rooms in an isolation ward, and 47 (32.6%) were admitted to cohorted bays in 2 medical wards.

Of the 144 patients available for analysis, 38 (26.4%) died during their hospital admission, 31 (21.5%) were admitted to ICU, and none had a cardiac arrest call. The duration of hospital stay ranged from 0.7 to 141.9 (median 8.9, IQR 5.6-19.8) days. The wearable monitoring period ranged from 0.15 to 440.2 (median 35.5, IQR 9.23-97.02) hours. Once a patient was placed on wearable monitoring, the median data availability was 88.5% (IQR 71.95%-98%). A total of 11,634 intermittent EWSs were recorded. During wearable monitoring, patients had continuous vital sign values that exceeded the EWS escalation threshold for 32.7% (2133/6528) of the total hours. The maximum intermittent EWS recorded was 17. The maximum continuous/partial EWS recorded was 9. The median maximum continuous/partial EWS for the duration of waves 1 and 2 was 3 (IQR 2-4) and 2 (IQR 1-4), respectively. Additional graphs showing the patterns of vital signs based on patient outcome and pandemic wave can be found in Multimedia Appendix 2.

For all data, the median increase in TTNO for monitored periods was 39 minutes (95% CI 29-49; P<.001; Table 1) longer than unmonitored periods. Figure 3 compares the distribution of TTNO values for these 2 periods, with the median values reported in Table 1 for the EWSs of all patients. Figure 3 shows that the 120-minute peak in TTNO for unmonitored periods was not present in monitored periods.

For stable periods (EWS of 0-2), the increase in median TTNO during monitored periods was 51 minutes (95% CI 39-62; P<.001) compared with unmonitored periods (Table 2). For unstable periods (EWS of ≥3), the increase in median TTNO during monitored periods was 16 minutes (95% CI 8-24; P<.001; Table 2).

The mixed effects model results for all patients (Table 3) show that both the use of wearable monitoring and unstable EWS periods were statistically significant (P<.001 for both terms) in association with TTNO. When adjusting for EWS category and participant-level clustering, the effect was attenuated but remained significant (14.6 minutes; P<.001). Re-estimating the model with EWS treated as a continuous variable showed similar results (P<.001 for both terms). Sensitivity analyses using gamma and log-normal mixed models showed consistent results in terms of direction and statistical significance. There was no evidence of interaction between clinical stability and monitoring period (P=.35), indicating that the effect of wearable monitoring on TTNO did not differ significantly between stable and unstable observations.

Figure 4 compares normalized score differences in EWS in the fourth quartile of TTNO for periods A (unmonitored periods: 275-898 minutes) and B (monitored periods: 326-830 minutes), demonstrating little difference between them (P=.92).

The prospective registration and the use of a rigorously developed unique WMS are strengths of this study. However, there are limitations that should be considered. Although we report a novel analysis that has the potential to advance our understanding of the impact of WMSs on nurse behavior, this analysis was post hoc and not included in our study registration.

The WMS was deployed under considerable pressure in the context of the COVID-19 pandemic and into a dynamic clinical setting including 3 different wards, which incurred several limitations. First, nurse behaviors during this time may have varied depending on the ward setting or evolved during the course of the pandemic and may not be comparable to conditions since the pandemic given the very high workload and stress at that time. Second, due to limited availability of equipment, nursing staff chose which patients to monitor using the system, which may have introduced selection bias as they were more likely to monitor those assessed at highest risk of deterioration. In addition, nurses may have taken fewer intermittent vital sign observations as patients recovered and neared discharge. However, discharge from hospital was accelerated during the pandemic due to resource pressures, which may have offered some mitigation against this potential confounder of the TTNO analyses. Third, the comparison between different time windows for the same patients could have been influenced by clinical staff who knew the patients by the time they were on the WMS. Thus, the observed pattern of monitoring might be more (or less) spaced given this potential bias and reflect the impact of competing clinical priorities. Fourth, the criteria for hospital admission and discharge evolved across the 2 pandemic waves, resulting in variability between patient clinical acuity at hospital admission and discharge, potentially confounding the TTNO analyses. Given the nature of this study, it was not possible to compare a completely nonmonitored cohort, which would have mitigated these first 4 limitations. Future research under nonpandemic conditions should consider using a separate unmonitored control group and setting specific eligibility criteria for commencing and ceasing monitoring to facilitate consistency of use. Fifth, although vital sign values from nurse-measured and wearable sources were compared, these 2 methods of measurement do incur differences, such as accuracy of devices and mode of measurement, which should be considered in the context of this analysis. In particular, nurse-measured respiratory rates may differ significantly from those measured using the WMS due to monitoring behaviors such as “digit preference” (where respiratory rate distribution shows peaks at 20 and 24 breaths per second). This suggests either estimation of respiratory rate based on observation or measuring over a shorter period of 15 seconds and multiplying by 4 [29,30]. Future research should consider protocolizing nurse-measured vital signs to ensure accuracy and avoid these behaviors.

We identified one limitation related to the WMS, which included a pulse oximeter device known to consistently underestimate oxygen saturation by up to 2%. However, the device was selected as the most reliable and accurate of those tested for the system [9], and notification of this underestimation was included in staff training and deemed by the nurses to be manageable and preferable to overestimation [26]. This may have resulted in nurses unnecessarily returning to patients to check their vital signs but did not provide false reassurance or risk masking hypoxia. Future studies should consider the impact of the comparative accuracy of devices on nurse monitoring behaviors.

There are some limitations to the data available. First, we planned to include additional adverse outcomes, including escalation to high-frequency nasal oxygen and noninvasive ventilation, in our analysis. However, these data were not available in the extracted dataset as patients undergoing these interventions were moved to a separate high-dependency ward. Future studies should ensure that data on these additional outcomes are available. Second, 5.7% (9/157) of the initial patients were not included in the analysis as they had registered for the NHS digital data opt-out service. Although this omission may pose a risk of selection bias, given the small number, this is unlikely to have impacted the results. Future research using electronic patient data will need to consider how missing data due to the opt-out service are handled if the proportion of patients choosing to opt out increases over time.

From a modeling perspective, TTNO was right skewed. Although we used a linear mixed effects model to preserve the interpretability of effect estimates as absolute differences in minutes, sensitivity analyses using alternative distributional assumptions (gamma and log-normal models) provided consistent results. However, departures from normality may have affected estimate precision.

Finally, our data were from a single hospital with a high percentage of ICU admissions (31/144, 21.5%). We acknowledge that these factors limit the generalizability of our findings. Future research should focus on exploring the impact of WMSs on nurse behaviors in nonpandemic conditions and comparing robust and distinct patient groups. Future consideration should also be given to the cost-benefit of the system given the impact on nurse behaviors and patient safety.

In this retrospective data study, patients were at high risk of deterioration, spending a third of the monitored time at or above the escalation threshold. Our post hoc analysis showed that, by offering additional vital sign data between intermittent EWS measurements, the WMS may have improved task prioritization by allowing the nursing staff to extend the time between routine MVSMs. We found no indication in our analysis that this increase in time between vital signs measurements adversely affected patient safety. The increase in time between observations was not associated with a difference in EWS distributions, providing some reassurance that this difference was not associated with increased physiological instability.