The COVID-19 pandemic stressed global health care systems, diverting resources from elective care and prioritizing the care of people with acute respiratory disease [1].

The introduction of a digital ward that commenced enrolling patients in November 2020 in Leicester, Leicestershire, and Rutland (LLR) was a key part of National Health Service (NHS) preparations for an expected surge in COVID-19 infections in the winter of 2020-2021.

The background to this decision was the impact of the first wave on LLR acute bed availability and the use of digital technology to remotely monitor patients to support their chronic conditions, both at the start of the COVID-19 pandemic and previously [2,3]. There was evidence to support the care of people after discharge with a high risk of readmission [2-4], and digitally supported patients with infectious respiratory disease had a reduced length of stay (LOS) [5-7].

The primary objectives for the COVID-19 digital ward were to maintain acute bed capacity through safe, early acute discharge and step-down into a specialist respiratory–managed service in the patients’ homes. Patients with COVID-19 respiratory infections were discharged into the care of the Leicester Partnership Trust (LPT) specialist respiratory team who were supported by digital technology (Clinitouch, Spirit Health), which electronically conveyed clinical observations and health status information to clinicians in an algorithm-based, traffic light (red, amber, and green)–prioritized basis. The standard operating procedures (service description), inclusion and exclusion criteria, objectives, and outcome measures for the digital ward are displayed on the internet [8], and more detail related to the intervention can be found in a study that regarded the first 65 patients that accessed the digital ward [9].

Paper-based, phone-based, digitally-based, and wearable device–based digital wards were used during COVID-19 as vehicles for admission avoidance and for stepping down patients who had been acutely admitted. There were mixed results [10].

The findings could have implications for future pandemics and how health systems allocate resources to better recover from the pandemic.

The objective of this study was to demonstrate the impact of a digital COVID-19 ward on NHS resource use. Specifically, the study aimed to establish if the digital ward achieved its primary goal of freeing up beds and the extent to which it reduced or increased overall NHS resource use.

There were 310 patients admitted to the University Hospitals Leicester (UHL) NHS Trust with COVID-19 respiratory disease; they were either discharged home between November 2020 and November 2021 into a digital ward to support their oxygen weaning (31/310, 10%) or had not required oxygen at discharge and needed additional support to recover (279/310, 90%). The patients’ mean age was 55.0 (SD 13.7; median 56; range 22-86) years, 3.2% (n=10) of patients were 80 years or older, and 40.6% (n=143) of patients were female. No ethnicity, comorbidity, or socioeconomic status information was collected. Patients were given the option to go home early and be supported digitally when they met the inclusion criteria [8].

This study was based on observational data and represents a cost minimization of a service evaluation. All patients discharged into the digital ward had confirmed COVID-19–related respiratory disease. The Consolidated Health Economic Evaluation Reporting Standards (CHEERS) 2022 guidelines for economic evaluations were followed [11] (see Multimedia Appendix 1). This cost minimization analysis only evaluated NHS resource use and the outcomes of interest were acute bed-days saved, their costs offset, and the costs of the digital ward. It was considered that the digital ward would not deliver any change in patients’ long-term, health-related quality of life or health outcomes.

The perspective taken was that of the English NHS. Any savings would have been in the acute sector and additional costs in the intermediate care sector. The resource use only considered the digital ward costs and the costs of its comparator, acute hospital care.

The time horizon was slightly more than over 12 months, and no discounting of costs was conducted. All costs were in 2020-2021 pounds sterling and converted to US dollars.

Comparator data for patients not on oxygen with acute LOS were sourced from patients discharged immediately prior to the introduction of the digital ward in November 2020 and a published NHS data set [12]. The difference between the duration of the LOS for patients discharged into the digital ward for oxygen weaning and those discharged routinely were acquired from UHL by LPT and NHS X (now part of the NHS Transformation Directorate). Mean clinical consultation durations and staff seniority were sourced from LPT (both correspondence: JS). Individual acute LOS, readmissions data, and the cost of a day in a respiratory ward were sourced from UHL (correspondence: SG). Staff unit costs were sourced from the Personal Social Services Research Unit (PSSRU) data set for 2020-2021 [13].

There were 2 principal costs associated with participants’ stay in the digital ward: the duration of patients’ digital ward LOS, which influenced the costs of the digital technology and was calculated on a per diem basis; and the number and duration of respiratory specialist–patient digital contacts. The duration of the digital ward stays and the number of contacts were sourced from the digital technology database. The method for estimating the mean duration of a clinical contact was described in the paragraph above.

Resource use data in both units and costs can be found in Multimedia Appendix 1.

There were 2 different populations discharged into the digital ward; those on oxygen at discharge and those who were not.

Those on oxygen were subjected to an analysis conducted by NHS X and LPT. They were discharged from the hospital, on average, 9.9 days earlier than similar patients who had not accessed the digital ward.

The first 65 patients not on oxygen discharged into the digital ward left acute care in 3.3 days, 2.2 days (40% relative reduction in LOS) earlier than controls who did not have the potential to access the digital ward [9]. The information on controls’ LOS (5.5 days) was sourced immediately prior to the digital ward’s introduction. This is an analysis of all patients admitted into the LLR COVID-19 digital ward prior to the end of November 2021 and prior to the availability of disease-modifying medicines that reduced acute COVID-19 illness severity. This study includes the first 65 patients.

The LOS in UHL was reported alongside the median and mean monthly LOS for COVID-19 discharges between March 20 and December 21 [12]. The linear relationship between UHL discharges and median acute LOS is illustrated in Figure 1. The red line displays the result of pulling the data on the median LOS (gray hashed line) 1 month back in time. In doing so, the r² improved from 0.31 to 0.75 in an ordinary least squares (OLS) model.

Mean LOS included a minority of patients with intensive resource use and very long LOS, skewing the LOS upward (see Figure 2 for differences between mean and median LOS for England and UHL). The estimated comparator LOS was calculated for patients who were discharged without requiring oxygen. The authors considered that the median LOS better reflected patients’ estimated LOS.

Figure 2 shows the mean and median LOS for England and for UHL [12]. The solid blue and orange lines show the mean and median English LOS. The hashed gray and yellow lines show the mean and median UHL LOS.

The 4 comparators below were then used to populate 2 simple simulations, creating 1000 random iterations for each of the 279 patients in both uniform and normal distributions, and the mean of both was taken as the base case.

The data model used the 4 comparators and was conducted in Microsoft Excel. The RANDBETWEEN function was used to create a uniform model with 1000 iterations between the lowest and highest values of the 4 comparators for each patient’s LOS and the mean of the 1000 iterations was taken for all 279 patients. The same process was conducted to create a normal distribution of those data using the NORMINV and RAND functions, using the mean of the 4 comparators and their SD to develop 1000 iterations, from which the mean was taken. The final mean was taken from the 2 mean parameters and provided the base case.

The NHS X and LPT analysis was used as the basis of the comparator for all patients treated with oxygen discharged into the digital ward.

A deterministic 1-way analysis was conducted to establish the extent to which varying the parameters influenced the savings. A 2-way analysis was developed to establish the extent to which the acute LOS and savings would have to be varied to reach the net savings threshold value of 0.

The data sets were not normally distributed for both study subgroups (patients on and not on oxygen) on acute discharge. To evaluate the data, 2-tailed Wilcoxon signed-ranks tests for paired samples were used. Individual patients’ LOS was compared with their comparators. Normal approximation and ties correction were used, and α was set to .05. These methods were used consistently to compare paired data.

A total of 55 pairs were required for a 90% power to establish a difference between the 5.5-day LOS (the lowest comparator) for patients not on oxygen. The scale of the difference between the patients who were oxygen weaning meant that the number of pairs required was only 4. There were 310 patients in the study: 279 (90%) patients who were not discharged on oxygen and 31 (10%) patients who were discharged home and into the digital ward for oxygen weaning.

Patients in the digital ward were monitored against a range of clinical criteria and classified by an algorithm into red, amber, and green ratings, which were highlighted on the LPT specialist respiratory clinicians’ dashboards. Red alerts provoked clinical contact within 24 hours throughout the patients’ stay in the digital ward; amber alerted the same, but in the first week only; and green rated patients who had not been contacted previously but were contacted in their second week in the digital ward. Patients were monitored using digital technology, which was billed on a per diem basis.

The study was evaluated by the Institutional Ethics Review body of DeMontfort University, with a decision reference of HLS FREC Ref: 2091/22. Approval was waived for the protocol as it was an economic analysis of a service that eligible patients routinely accessed as part of standard care. All analyzed data were deidentified. All patients accessed usual care. No payments were made to authors or patients. Patients or the public were not involved in the design, conduct, reporting, or dissemination plans of our research.

All analyses were conducted in Microsoft Office Excel, Excel Analysis Toolpak, and The R Foundation’s R Studio.

The 2 different patient populations had differing acute LOS. For patients not on oxygen when discharged, the mean acute LOS was 4.2 (SD 2.1; median 4.2) days. For patients treated with oxygen, the mean acute LOS was 13.3 (SD 4.5; median 12.4) days. Overall, the mean acute LOS was 4.3 (SD 2.1; median 4.4) days.

The phasing of patients discharged into the digital ward reflected demand in UHL [14]. The number of discharges into the COVID-19 ward (bars in Figure 3) corresponded with key system stressors (lines in Figure 3) rising and falling. The 3 key stressors reported were the ratios of COVID-19–related absences divided by all absences, total COVID-19–related beds occupied divided by all beds occupied, and COVID-19–related mechanically ventilated beds divided by all mechanically ventilated beds occupied [14].

Table 1 demonstrates the difference in acute LOS versus the imputed non–oxygen weaning LOS comparators.

The mean of the normal and uniform distributions from 1000 random iterations of the above data in each of the 279 non–oxygen weaning comparators were 6.31 and 5.95, respectively. The mean of both, 6.13, was used as the base case.

A UK study of patients with COVID-19 not admitted to an intensive care unit estimated the acute LOS as 8.0 to 9.1 days [15]. All acute LOS comparators in the non–oxygen weaning group fell below this range, with the base case being almost 2 days beneath the lower value [15], supporting the conservatism of the model. The mean digital ward LOS was 14.2 days (SD 4.9; median 15) for patients not on oxygen.

The mean acute LOS for patients discharged into the digital ward on oxygen was 13.3 days, and the difference was 9.9 days (42.7%) between the actual LOS and the comparator LOS (P<.001). The mean digital ward LOS for patients on oxygen was 24.5 (SD 12.6; median 23) days.

Digital ward resource use was driven by 2 elements: the number of clinical contacts and the LOS; the details are shown in Table 2. Multimedia Appendix 2 contains a breakdown of unit costs. How they were calculated is outlined in the Methods section.

The range of estimated bed-days saved in the 279 patients not on oxygen varied from 363.8 (constant 5.5 days) to 749.9 (median OLS+10%); the comparison of digital ward patients versus all imputed controls was significant (P<.001). The base case was estimated to have saved 490.3 bed-days in the patients not on oxygen (P<.001). The estimated bed-days saved in the 31 patients on oxygen was 306.9, with which there was greater certainty (P<.001). The total estimated bed-days saved were between 670.7 and 1056.8 (2.2-3.4 days per patient). The base case estimate was 846 bed-days saved.

The cost of a bed day in a UHL acute respiratory ward in November 2020 was US $678.

Estimated implications on costs are summarized below:

There were 9 hospital readmissions (9/310, 2.9% of digital ward admissions) within 30 days. All readmissions were in the non–oxygen weaning cohort. In a systematic review, 10.3% (n=265,590) of those admitted with COVID-19 respiratory infections had a 30-day readmission [16]. Of the included English studies, 3 reported on 30-day readmission rates but included patients who had accessed critical care or had a high percentage of older adult patients [17-19] and reported readmission rates of 10.2% to 17.1%.

A total of 7.1% (n=154) of patients with mild or moderate COVID-19 disease, similar to the digital ward patients not on oxygen, discharged from a Turkish tertiary center were readmitted within 30 days [20]. Another systematic review [21] found that readmissions ranged from 4.2% [22] to 19.9% [23].

Evidence suggested that there was no difference between the rates of 30-day readmissions between the first and second COVID-19 waves [24]. If it was assumed the readmission rate for a comparator was 7.1% [20], the costs of readmissions would have been US $69,050 or 99.6% (US $69,293) of the total costs of running the digital ward. Using the lowest plausible comparator of 4.2% [22] would have offset 30.8% (US $21,342) of the costs of the digital ward. Any potential savings associated with readmissions were excluded from the results of the analysis because of the uncertainty associated with any potential comparator and to aid in the overall conservatism of this study.

The carbon footprint associated with acutely hospitalized patients has been described as the most carbon-intensive care pathway and contributed 125 kg of carbon dioxide equivalent per day [25]. The gross reduction in 2019 carbon dioxide equivalent was 341 kg per patient in the base case and totaled 105.75 metric tonnes.

There were 8 variables in this analysis; the differences in acute LOS for those who were on the digital ward versus comparators, the cost of an acute bed-day, the number of clinical contacts, the clinical contact duration, the cost per hour of clinicians, the digital ward LOS, and the digital technology costs.

A 1-way sensitivity analysis demonstrated that the 2 variables that changed by more than the relative input parameters were the LOS of patients who were and were not oxygen weaning. There was a linear 3.18% and 2.34% increase or decrease in the savings or costs for every 1% change in the value for the patients who did or did not access oxygen, respectively. All other resource-use variables increased or decreased at the same rate as the input variable. The parameter with the greatest uncertainty was the one that had the greatest impact on potential savings.

The cost of clinical consultations (respiratory specialists, LPT) and the cost of digital technology (Clinitouch, Spirit Health) were the 2 components of the digital ward costs. When the costs of both were simultaneously increased in a 2-way sensitivity analysis in the 279 patients not on oxygen by the same percentage as the LOS was reduced, it required a 75.4% change to reach the 0-threshold value. Performing the same for patients who accessed remote oxygen weaning, the threshold value was reached by simultaneously increasing costs and reducing the LOS cost offset by 87.5%. Overall, it took a reduction in effect and increase in costs of 78.4% (savings reduced from their estimated base case value of US $573,490 to 21.6% of that US $123,874 and costs to rise from US $69,293 to US $123,618) to reach a 0-threshold value.