Budget impact analysis of providing hospital inpatient care at home virtually, starting with two specific surgical patient groups.

OBJECTIVE: To determine the budget impact of virtual care.
METHODS: We conducted a budget impact analysis of virtual care from the perspective of a large teaching hospital in the Netherlands. Virtual care included remote monitoring of vital signs and three daily remote contacts. Net budget impact over 5 years and net costs per patient per day (costs/patient/day) were calculated for different scenarios: implementation in one ward, in two different wards, in the entire hospital, and in multiple hospitals. Sensitivity analyses included best-case and worst-case scenarios, and reducing the frequency of daily remote contacts.
RESULTS: Net budget impact over 5 years was €2 090 000 for implementation in one ward, €410 000 for two wards and €-6 206 000 for the entire hospital. Costs/patient/day in the first year were €303 for implementation in one ward, €94 for two wards and €11 for the entire hospital, decreasing in subsequent years to a mean of €259 (SD=€72), €17 (SD=€10) and €-55 (SD=€44), respectively. Projecting implementation in every Dutch hospital resulted in a net budget impact over 5 years of €-445 698 500. For this scenario, costs/patient/day decreased to €-37 in the first year, and to €54 in subsequent years in the base case.
CONCLUSIONS: With present cost levels, virtual care only saves money if it is deployed at sufficient scale or if it can be designed such that the active involvement of health professionals is minimised. Taking a greenfield approach, involving larger numbers of hospitals, further decreases costs compared with implementing virtual care in one hospital alone. © Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

We deconstructed the cost of hospital inpatient days to more accurately estimate potential cost savings.

As fixed costs constitute a major component of the cost of hospital inpatient days, we used capacity estimation to assess possible reductions in fixed costs.

We explored the effect of various levels of scale on the estimated budget impact.

Many assumptions were made, owing to the novelty of the conceived intervention, and a consequent lack of an evidence base.

This study was conducted within the context of the Dutch healthcare system, which may limit the generalisability of the results.

Background

Healthcare costs have been rising for decades and are expected to increase even further. Hospital care expenditure comprised more than 30% of total healthcare costs in the USA and in 29 out of 31 countries in the European Economic Area.1 2 To reduce the cost growth of hospital care, attempts are being undertaken to move care for postoperative and comparable patient categories out of hospitals to lower-cost contexts, such as the home situation and primary care.3–7 Most of these attempts involve provision of in-person care in patients’ homes by medical specialists. As a result, medical specialists spend much time on travel between patients. Use of digital technologies may allow more efficient use of healthcare resources, by entirely eliminating travel time and by enabling nurses to carry out most of the work. One way to move care out of the hospital using digital technology is through telehealth, defined in a systematic review as ‘the use of information or communication technology as a medium for enabling professional–patient interaction’.8 Telehealth includes monitoring patients remotely, by telephone calls, store and forward services, or automatic monitoring devices enabling detection of patient deterioration, as well as teleconsultations or videoconsultations, websites, or smartphone apps to provide health advice to patients.

While manufacturers frequently claim that telehealth reduces the use of hospital services and generates cost savings, thorough evidence for this is lacking.9 10 Properly designed studies are rare both due to technology push and rapid development of innovative technologies. While the use of telehealth to manage chronic conditions such as heart failure and chronic obstructive pulmonary disease (COPD) is well studied and is generally positive,11–13 these patient categories are at relatively low risk of complications requiring hospitalisation compared with postsurgery patients. Presently, hospitals in the Netherlands are increasingly investing in telehealth to substitute in-hospital care. These investments are typically made by individual hospitals or departments within hospitals. This may not be the most cost-efficient way to organise telehealth, however, the effect of scale on telehealth costs has not been studied thoroughly. Furthermore, investments in telehealth are often predicated on the idea that cost savings will be achieved. As a result, payors intend to lower hospital budgets, despite a lack of evidence.

The case we present is a first step in developing a virtual care setting for hospital patients, using remote monitoring to enable very early discharge of postoperative and comparable categories of patients who need frequent supervision, and who would usually remain in hospital for observation for at least 1 day. The virtual care centre enables patients to be monitored in their homes, aiming to reduce the number of hospital inpatient days. This article presents a budget impact analysis (BIA) of a case at a large teaching hospital in the Netherlands. We consider four scenarios for organising virtual care: (1) implementation of virtual care in a single ward, (2) virtual care in two wards, (3) providing virtual care in an entire 766 bed hospital through a hospital-based virtual care centre and (4) providing virtual care for all Dutch hospitals, that is, 39 900 beds, through a ‘greenfield’ approach.

Methods

Patient and public involvement

Patients or the public were not involved in this study. Dissemination to participants or patient groups is not applicable.

We conducted a BIA of virtual care from a hospital perspective with a time horizon of 5 years, using a cost-calculator approach following International Society for Pharmacoeconomics and Outcomes Research (ISPOR) guidelines.14 Virtual care includes a wireless wearable sensor that continuously measures vital signs, a relay device that transmits measurements to the hospital and a number of teleconsultations or videoconsultations. Online supplemental appendix A describes virtual care in more detail.

Costs taken into account, the four scenarios, and the assumptions made in the calculations are provided below.

Cost types

To provide insight into how savings could be achieved, costs are separated into investments, fixed costs and variable costs. Investments are those costs necessary to enable virtual care that only vary with the maximum number of patients expected to receive the intervention. Fixed costs are not directly affected by variation in the number of hospital bed days provided to patients or the number of patients, but may be reduced if hospital bed days are reduced by a sufficient amount. Variable costs vary directly with the number of bed days or the number of patients. Costs are further subdivided into costs related to (1) technology, (2) infrastructure, (3) service, (4) start-up and (5) inpatient days. Table 1 provides a complete overview of costs.

Table 1

Overview of costs

	Investments		Fixed costs		Variable costs
	Cost item	Price	Cost item	Price	Cost item	Price
Technology	Relay device	€1150			Biosensor	€120
	Patient licence	€520
	Client licence	€130
	Mobile client licence	€170
Infrastructure	Server hardware	€33 900	Offices*	€1200
	Software solution	€12 100
	Access point	€1390
Service			Telenurse	€65 000	Home visit	€80
			Remote Technical Support	€16 000	Home treatment	€130
					Ambulance transport	€760
Start-up	Project management	€48 400
	Technical implementation	€20 000
	External consultancy	€40 500
	Education	€25 000
Inpatient day			Specialists	€30	Materials	€10
			Physician’s assistants	€20
			Nurses	€220	Room and board	€70
			Real estate	€20
			Overhead	€130

Cost is per m2 per year. All other prices are unit prices and include 21% Value Added Tax (VAT).

Investments

Investments need to be made in technology and infrastructure, as well as in start-up costs, including implementation. For the technology component, investments include relay devices, client licences, mobile client licences and patient licences. Relay devices and patient licences are needed for each patient that is concurrently monitored with a sensor. A small reserve of relay devices may be needed, as they have to be returned by or picked up from the patient. Client licences are required for each access point in the remote monitoring centre. Mobile client licences are necessary for each mobile device with access to the server. All infrastructure costs are investments, that is, server hardware, software licence fees and access points consisting of computers with monitors, tablet computers to enable videoconferencing and office furniture. The server is capable of monitoring 240 sensors simultaneously.

All start-up costs are depreciated as investments. These arise from project management to guide implementation, technical implementation to integrate the new technology and all of its components into existing systems such as the electronic medical record and ensuring system security and compatibility, external consultancy for various purposes, training nurses in using the new equipment, as well as training telenurses that staff the virtual care centre.

Fixed costs

Fixed costs originate from the infrastructure, service and hospital inpatient capacity components and are also related to the offices necessary for the remote patient monitoring centre. Fixed costs of the service include costs of telenurses and costs of remote technical support, which enables the vendor of the sensors and software applications to intervene if necessary. The major part of costs for inpatient days comes from salaries for specialists, physician assistants and nurses, as well as real estate and overhead. These costs are reported as costs per inpatient day, as this is how they are conventionally quoted and reimbursed. In reality, however, in many countries, these costs are fixed on the short-term and mid-term rather than variable.

Variable costs

Costs for the technology, service and inpatient days contain variable components. In the case of technology, only the costs of sensors are variable, as patients need their own sensor.

As described in box 1, some patients may require a home visit, home treatment or ambulance transportation to the hospital, resulting in a variable cost component. Finally, a small proportion of costs for inpatient days is variable, consisting of materials such as medication, bandages, office supplies, and room and board. To estimate changes in nurse costs, capacity estimation is performed (online supplemental appendix A) based on a method developed in a different study.15

The virtual care centre is meant to facilitate very early discharge of patients from the hospital, thereby reducing the number of inpatient days. It consists of three main components: (1) technology, (2) infrastructure and (3) service. Each component is described below.

Technology

The technology component consists of a wireless wearable sensor, referred to as ‘biosensor’ and a relay device. The biosensor continuously measures patients’ health status in terms of respiratory rate, heart rate, heart rate variability, skin temperature and body posture. It is able to do so for 4 days (96 hours). The relay device receives the biosensor data through Blue Tooth Low Energy and transmits the data to the hospital through wireless internet.

Infrastructure

The infrastructure consists of server hardware and a software solution to process the biosensor data, and a remote monitoring centre. The server hardware must be powerful enough to process a large continuous stream of data from several biosensors simultaneously. The software solution processes the data and provides a comprehensible overview of patients’ health status. It is also capable of generating automated alerts.

The hospital-based virtual care centre is equipped with a number of access points to the software solution, enabling simultaneous monitoring of all patients who are wearing a biosensor, as well as inspection of the complete biosensor measurement history. It is staffed by specially trained telenurses, each of which requires an access point.

Service

On discharge, patients are equipped with a biosensor and a relay device. Telenurses contact patients at least three times daily by telephone or videoconferencing for a duration of 4 days. Based on the assessment of patient health status, telenurses decide whether to provide behavioral or medication advice, to conduct a home visit or home treatment, or to contact a specialist. In case of a home visit or home treatment, it is desirable from a practical standpoint that another nurse is available as back-up, meaning there must always be at least two telenurses, regardless of how many patients are under the care of the virtual care centre. If a specialist is contacted, they may determine that immediate transfer to the hospital by ambulance is desirable, or otherwise may give instructions to transfer a patient to hospital if symptoms progress to a certain point.

Data sources

Costs of technology, server hardware, the software solution, remote technical support, technical implementation and education are based on a quotation of the vendor of the telehealth intervention. Costs for access points are based on market prices for equipment currently in use. Project management and external consultancy costs were retrieved from internal documents of the hospital. Costs of inpatient days are based on 2014 weighted average reference prices of general and academic hospitals, retrieved from online supplemental material to the Dutch guideline for economic evaluations in healthcare.16 After correcting to 2019 values by applying the Consumer Price Index,17 these prices were used for the calculations. Telenurses will be responsible for a large number of patients, require a broader knowledge base to deal with a large variety of health conditions, and need to be able to provide care at a distance. Therefore, costs are expected to be higher than for a conventional nurse, but lower than for a nurse working in an intensive care unit. Hospital admissions data needed for capacity estimation were acquired from the hospital’s business intelligence department.

Scenarios

The strategy of establishing a hospital-based virtual care centre will be explored through four scenarios in which the expected effect is modelled on hospital admissions data from 2015 to 2019. Box 1 presents the details of the virtual care centre. A full description of each scenario is presented below. Because the different scenarios may hinder the comparability of budget impact figures, we additionally present costs of virtual care per patient per day.

Scenario 1: single ward - bariatric surgery

The bariatrics ward is a 16 bed ward. From 2015 to 2019, bariatric surgery was performed in 1295 (SD=72) unique patients per year, who used an average of 3897 (SD=202) bed days. Additionally, 380 (SD=91) non-bariatric surgery patients used this ward per year. The average length of stay was approximately 2.5 days.

Patients who undergo surgery in the morning are typically discharged in the afternoon of the next day. With virtual care, bariatric surgeons expect that these patients could be discharged in the evening of the day of surgery, as long as they meet the following criteria: (1) being free from diabetes or sleep apnea, (2) living within 30 min of the hospital (by car), (3) not living alone and (4) they or a caregiver are capable of working with the technology. It is assumed that surgeries are planned in such a way that all patients undergoing surgery in the morning meet eligibility criteria for same-day discharge. Based on this, an average of 402 patients (SD=83) would have been eligible for virtual care per year.

Scenario 2: two wards and different patient groups - bariatric and vascular surgery

The vascular ward is a 19 bed ward, which provided services to 927 (SD=63) unique patients per year from 2015 to 2019, on average. Based on expert opinion, it is expected that patients treated for carotid artery pathology (mean N=71, SD=6) or endovascular treated abdominal aortic aneurysm (AAA; mean N=156, SD=18) are eligible for very early discharge with virtual care. The average length of stay for these patients was approximately 3.2 and 8 days, respectively. For carotid artery surgery a prolonged hospital stay is due to postoperative hypertension, and for AAA patients due to postoperative fever. Vascular surgery patients must meet the same criteria as bariatric surgery patients to be eligible. We assume again that the planning can be made such that all patients undergoing surgery in the morning meet eligibility criteria. Based on this, an average of 196 patients (SD=21) would have been eligible for virtual care per year.

Additional investments in technology will be needed, as an increased number of patients leads to the need for a greater number of relays and patient licences, and may cause a greater number of client licences and mobile client licences to be necessary.

Scenario 3: hospital-wide implementation in one hospital

The hospital in this case had 766 active beds from 2015 to 2019, and provided care to 33 295 (SD=427) unique patients per year. We calculated the weighted average of the proportions of eligible patients found in scenarios 1 and 2 for each year, resulting in an eligible proportion ranging from 19% to 32% of all patients. Thus, an average of 8517 (SD=1640) patients would have been eligible for virtual care in the whole hospital per year. A weighted average is also calculated for the number of days by which length of stay is reduced, to determine the total number of inpatient days that can be saved in this scenario.

In scenarios 1 and 2, the number of inpatient days to be saved to reduce the number of nurse shifts by one is calculated. Based on this, a weighted average is calculated. This is done by dividing the number of inpatient days that can be saved by the number of inpatient days to be saved to reduce the number of nurse shifts by one. As the wards studied in scenarios 1 and 2 turned out to be relatively close to being able to reduce the number of nurse shifts, we further took the average of the theoretical production per nurse shift and the weighted average (Online Supplemental File 1) to produce a less optimistic estimate for the base case.

Scenario 4: multiple hospitals - greenfield

In this scenario, a virtual care centre is established independent of any one hospital, which provides its service to a number of hospitals, in our case the whole of the Netherlands. The proportion of patients who are eligible and the reduction in length of stay are based on findings from the first two scenarios, as for scenario 3. To account for differences in hospital size, we calculated the average number of patients receiving virtual care per bed per year in scenario 3, which we multiplied by the total number of hospital beds in the Netherlands in 2018 (N=39 900)18 to arrive at the number of patients receiving virtual care in all Dutch hospitals combined per year.

It is assumed that investments in technology, infrastructure and start-up costs are needed once per hospital.

General assumptions

Besides the assumptions described in the scenarios above, some general assumptions were made which may differ in other healthcare systems, namely: (1) the hospital cannot increase its revenue by performing more surgeries per day, as health insurers impose volume restrictions on all Diagnosis Related Groups (DRGs),19 (2) capacity that becomes available due to early discharge of patients with the sensor is not used by patients from other wards or hospitals, (3) health outcomes do not change as a result of virtual care and (4) there is no impact on overhead costs. It should be noted that the volume restrictions per DRG can be renegotiated. However, as health insurers are tasked with keeping costs low, it is nevertheless questionable to what extent increased revenues could be achieved. An overview of all assumptions made is provided in box 2.

The hospital cannot increase its revenue by performing more surgeries per day, due to restrictions imposed by health insurers.

Capacity that becomes available due to early discharge of patients with the biosensor is not used by patients from other wards or hospitals.

There is no impact on overhead costs.

Reductions in nurse shifts are possible in increments of 0.5 shifts.

Nurses are each responsible for four beds during the day, six in the evening, and ten during the night.

Surgeries for eligible patients can be planned in the morning.

All patients treated for abdominal aortic aneurysm (AAA) or carotid artery pathology are eligible for virtual care.

All surgeries for AAA or carotid artery pathology can be planned in the morning.

Proportions of patients eligible for virtual care in scenarios 1 and 2 translate linearly to hospital-wide scale.

Reductions in nurse shifts are linearly related to reductions in hospital bed days.

Other hospitals are similar in size to the case hospital.

Sensitivity analyses

For scenarios 3 and 4, it is investigated to what extent the budget impact would be affected by changes in the proportion of eligible participants, the reduction in length of stay, and the number of telephone contacts that are performed as part of virtual care. Additionally, the effect of allowing repurposing of saved inpatient days such that they can be utilised by patients from other wards or hospitals is explored.

Costs per patient per day and net budget impact are calculated for a best-case and worst-case scenario. In the best-case scenario, the greatest proportion of eligible patients as well as the greatest reduction in length of hospital stay of the first two scenarios is taken, rather than the weighted average. For the worst-case scenario, the smallest value is taken for both. The effect of changing the number of telephone contacts provided to virtual care patients after discharge from three per day to one per day is also examined, which changes the ratio of telenurses to patients from 1:12 to 1:36. The effect of repurposing capacity is calculated for 20%, 50% and 80% of saved inpatient days. Capacity that is repurposed directly leads to savings of the total costs of an inpatient day, ie, €500 per day saved (table 1). As capacity is intended to be used in these cases, savings from reducing nurse shifts do not apply.

Validation

Face validity of the calculations was verified through discussions with a major health insurer in the region, industry partners and within the hospital with the financial director and business controllers.

Results

This section presents the budget impact over 5 years, as well as the cost of virtual care per patient per day for each scenario.

Scenario 1: single ward

If virtual care was implemented in a single ward, the number of nurse shifts during the day and the evening could be reduced by 0.5 shifts. Night shifts could not be reduced. The net budget impact is estimated at an additional €580 000 in the first year, followed by €377 500 (SD=€10 900) in subsequent years, resulting in a total net budget impact of €2 090 000 over a period of 5 years. As shown in figure 1, the majority of additional costs is caused by the service component of virtual care. Variable inpatient day costs provide average savings of €42 320 (SD=€15 850) per year and reductions in nurse shifts provide savings of €116 600. Net costs per patient per day are €303 in the first year and €259 (SD=72) in subsequent years.

Figure 1

Budget impact per year for scenarios 1 and 2.

Scenario 2: two wards

Simulating virtual care in two wards, the number of nurse day and evening shifts needed is reduced by 1.5 each, while the number of night shifts needed is reduced by 0.5. This results in cost savings of €419 760 in fixed costs. The net budget impact is estimated at an additional €262 000 in the first year, followed by €37 500 (SD=€28 700) in subsequent years, resulting in a total net budget impact of €410 000 over a period of 5 years. As shown in figure 1, the majority of additional costs is caused by the service component of virtual care. Net costs per patient per day are €94 in the first year and €17 (SD=€10) in subsequent years.

Scenario 3: hospital-wide implementation

Extrapolating the results of the effect of virtual care on capacity to the entire hospital shows that the number of nurse day shifts needed is reduced by 17.1 (SD=3.1), evening shifts by 12.9 (SD=2.4) and night shifts by 7.7 (SD=1.5) per year. This results in average cost savings of €4 531 000 per year in terms of fixed inpatient day costs.

The net budget impact is estimated at €474 500 in the first year, followed by €−1 670 000 (SD=€1 249 500) in subsequent years, resulting in a total net budget impact of €−6 206 000 over a period of 5 years. Net costs per patient per day are €11.1 in the first year, and €−55 (SD=€44) in subsequent years.

Scenario 4: all Dutch hospitals

Implementing virtual care in all Dutch hospitals, the number of nurse day shifts could be reduced by 933, evening shifts by 702, and night shifts by 423. This results in cost savings of €247 457 000 per year in terms of fixed inpatient day costs.

The net budget impact is estimated at €−65 824 500 in the first year, followed by €−94 968 500 in subsequent years, resulting in a total net budget impact of €−445 698 500 over a period of 5 years, providing virtual care to 2 218 045 patients, for a total of 8 872 180 virtual inpatient days. The macro impact for the Netherlands would thus be −0.1%. Net costs per patient per day are €−37 in the first year, followed by €−54 in subsequent years.

An overview of the results from sensitivity analyses is provided in table 2 and table 3.

Table 2

Sensitivity analysis for implementation of virtual care in a single hospital

		Year 1	Year 2	Year 3	Year 4	Year 5
Net budget impact	Base case	€474 355	€−2 823 290	€−1 342 786	€−2020	€−2 462 412
	20% Capacity repurposed	€4 569 241	€2 397 105	€3 259 670	€3 030 186	€1 680 108
	50% Capacity repurposed	€1 611 687	€−1 331 605	€−52 075	€805 791	€−1 331 518
	80% Capacity repurposed	€−1 345 866	€−5 060 315	€−3 363 820	€−1 418 603	€−4 343 144
	Worst case	€2 542 258	€2 164 298	€2 151 054	€2 157 333	€2 156 383
	Best case	€−9 692 469	€−7 864 258	€−8 482 822	€−7 410 863	€−5 811 811
	One telephone contact	€−2 351 096	€−5 046 923	€−3 894 347	€−1 989 368	€−4 269 629
Costs/patient/day	Base case	€1105	€−8383	€−3627	€−165	€−9762
	20% Capacity repurposed	€10 647	€7117	€8804	€ 9625	€6661
	50% Capacity repurposed	€3755	€−3954	€−141	€2559	€−5279
	80% Capacity repurposed	€−3136	€−15 025	€−9086	€−4506	€−17 218
	Worst case	€9979	€8417	€8658	€8531	€8560
	Best case	€−22 585	€−23 350	€−22 912	€ −23 539	€−23 041
	One telephone contact	€−5478	€−14 985	€−10 518	€ −6319	€−16 927

Table 3

Sensitivity analysis for implementation of virtual care in all 69 Dutch hospitals

	Net budget impact year 1	Net budget impact years 2–5	Cost/patient/day Year 1	Cost/patient/day Years 2–5
Base case	€−65 824 447	€−94 968 481	€−3710	€−5352
20% Capacity repurposed	€165 462 691	€136 318 657	€9325	€7682
50% Capacity repurposed	€2 526 775	€−26 617 259	€142	€−1500
80% Capacity repurposed	€−160 409 142	€−189 553 176	€−9040	€−10 682
Worst case	€122 094 790	€96 823 446	€9264	€7346
Best case	€−505 760 201	€−538 675 005	€−22 783	€−24 266
One telephone contact	€−184 881 086	€−213 262 360	€−10 419	€−12 019

If capacity could be repurposed so that it could be used by patients from other wards or other hospitals, costs of virtual care per patient per day in the first year for one hospital ranged from €107 for 20% capacity repurposed to €−31 for 80% capacity repurposed. In subsequent years, this range was from €81 (SD=€14) to €114.6 (SD=€57.7). For implementation in all hospitals, costs of virtual care per patient per day ranged from €93 to €−90 in the first year, and from €77 to €−107 in subsequent years. The net budget impact in one hospital ranged from €4 569 000 to €−1 346 000 in the first year, and from €2 592 000 (SD=€709 000) to €−3 546 500 (SD=€1 578 000) in subsequent years. For all hospitals, the net budget impact ranged from €165 462 500 to €−160 409 000 in the first year, and from €136 318 500 to €−189 553 000 in subsequent years.

In the worst-case scenario, only 19% of patients are eligible and length of hospital stay is reduced by 1 day per patient. The costs of virtual care per patient per day in the Netherlands were €100 for implementation in one hospital and €93 for implementation in all 69 Dutch hospitals in the first year. In subsequent years, costs per patient per day were €85 (SD=€1) for implementation in one hospital, and €74 for implementation in all Dutch hospitals. The net budget impact was €€2 542 000 for one hospital and €122 095 000 for all hospitals in the first year. Net budget impact in subsequent years was €2 157 500 for one hospital and €96 823 500 for all hospitals.

In the best-case scenario, 32% of patients are eligible and length of hospital stay is reduced by 3 days per patient. The cost of virtual care per patient per day in the Netherlands in the first year ranged from €−226 to €−228 for all hospitals, and from €−232 (SD=€3) to €−243 in subsequent years. The net budget impact ranges from €−9 692 500 for one hospital to €−505 760 000 for all hospitals in the first year, and from €−7 392 500 (SD=€1 141 500) for one hospital to €−538 675 000 for 78 hospitals in subsequent years.

When the number of telephone contacts per day in virtual care is reduced to one instead of three, the cost per patient per day in the Netherlands ranges from €−55 to €−104 in the first year, and from €−122 (SD=€47) to €−120 in subsequent years. The net budget impact ranges from €−2 351 000 for one hospital to €−184 881 000 for all hospitals in the first year, and from €−3 800 000 (SD=€1 299 000) for one hospital to €−213 262 500 for all hospitals in subsequent years.

Discussion

Replacing in-hospital care with virtual care does not directly lead to cost savings. This is mainly due to the fact that the virtual care service, in this case remote vital signs monitoring with three daily telephone contacts, apart from investments in IT infrastructure, requires additional telenurses to be available 24/7. Despite reductions of in-hospital nurse shifts, the additional costs of telenurses outweigh the savings made, until sufficient scale is reached. Although the two wards that were used as a basis for the analysis showed a level of utilisation that enabled nurse shifts to be reduced relatively quickly, this was insufficient to result in cost savings. In the base case, we start seeing savings over a period of 5 years once virtual care is implemented in the whole hospital. Employing a greenfield strategy, in this case involving all Dutch hospitals, does not show a much lower cost per patient per day than implementing virtual care in one hospital (tables 2 and 3). This indicates that the cost floor can be reached with fewer participating hospitals. In the best-case scenario, if virtual care is implemented in all Dutch hospitals, cost savings would be €538 675 000 per year. These cost savings will not reduce healthcare expenditure with considerable impact, though, as the macro impact would be approximately −0.5% of the total healthcare expenditure of the Netherlands. It must be noted, however, that we did not include indirect costs such as coordinating virtual care between all hospitals, relationship management or redesign of healthcare pathways. Especially when implemented at the scale of an entire country, these costs may have a significant impact on the savings that can be realised, which could reduce the macro impact to below −0.5%. As such, if the goal is to save money, it is questionable whether this is the approach that should be taken. Since the approximately 1 million bed days that could be saved under the assumptions in the base case equate to saving around 4000 beds (at 70% utilisation), it may instead be more interesting to consider virtual care as a way of increasing hospital capacity at relatively low cost. Our findings contradict those of communications from industry and prior research, which often found cost savings, also at smaller scales. In fact, only one prior study, also conducted in the Netherlands, was found to report an increase in costs.20 Studies reporting cost savings often assumed that inpatient hospital day costs are entirely variable,21–23 which is in contrast to the finding in this study that inpatient hospital day costs consist of a fixed component (84%) and a variable component (16%). Additionally, some studies reporting cost savings did not account for costs related to the intervention,9 11 while even implementation of virtual care in only one ward costs €728 000 in this study. Lastly, it is important to note that the case mix of diagnoses per hospital does not easily allow for a general or large scale reduction of staffs and wards, as digital services also require dedicated infrastructure and staffing; reductions mostly need to be specified per patient group.

This study must be interpreted within the context of several assumptions. First, it is not known with certainty which or how many patients are eligible for early discharge, as this is not common practice. Two eligible patient groups from different wards were identified based on expert opinion. The proportions of eligible patients relative to the total number of patients treated in these wards informed our analysis (19%–32% eligibility). Moreover, it is assumed that remote continuous monitoring devices combined with telephone or video contacts is equivalent to in-hospital care in terms of health outcomes. While health outcomes in chronic conditions such as heart failure and COPD are generally affected positively,24 25 little research has been done in directly postoperative or comparable patient populations from a case mix perspective. Nevertheless, the intervention in this case is of such short duration that improved health outcomes cannot realistically be expected. Furthermore, the greenfield analysis is a linear extrapolation of the findings based on data of a large teaching hospital. In reality, the results may well differ for other types of hospitals, as differences in the number of patients treated per day and length of stay between hospitals were not taken into account. Our hospital is, however, one of the larger Dutch teaching hospitals and we have no reason to believe admission patterns are very different in other hospitals. Furthermore, in the Dutch healthcare system hospitals cannot freely increase the number of patients that are treated, as health insurers impose volume restrictions on all DRGs. To assess the impact of this assumption on the results, we conducted a sensitivity analysis on the base case, where 20%, 50% or 80% of saved inpatient days could be repurposed to lead to cost savings directly, rather than through reducing the number of nurse shifts needed. This analysis showed that being able to repurpose 80% of inpatient days saved would result in a lower net budget impact than the base case. Setting this parameter to 50% instead results in a greater net budget impact than the base case. Break-even between the two is likely to be somewhere in the middle between 50% and 80%. Finally, in the Dutch implementation of DRGs hospitals are reimbursed based on a category of length of stay, rather than being reimbursed for every actual inpatient day. Examples of categories are <5 days, 5–10 days, 11–25 days and >25 days. If a patient moves from the category 5–10 days to <5 days, savings in in-patient days may thus lead to reduced hospital income.19 Although there are few examples of successful virtual hospitals and their definitions and scope vary per health system, different financing and market environments may lead to different degrees of impact.

This study also has several strengths. First, the cost of inpatient days was deconstructed to determine to what extent it consists of variable costs. Second, we used capacity calculation to establish the amount of fixed costs that could be saved. Third, we explored the effect of various levels of scale on budget impact.

An important implication of the results of this study is that it is essential for the success of virtual care’s potential for cost savings that it is implemented at sufficient scale. Furthermore, it is notable that limiting active involvement of health professionals in virtual care also makes cost savings more achievable. If the monitoring process can be automated, and health professionals need only take action when there is clinical necessity, costs can be reduced by an amount comparable to repurposing 80% of saved inpatient days. Validated algorithms which can detect or even predict deterioration in patients' health status must therefore be developed. Another possibly interesting avenue for future research is to investigate how early discharge can affect waiting lists, as well as optimise throughput from the emergency department and intensive care units to general and specialty wards through improved bed availability. Finally, it is important to consider that with virtual care, health professionals are responsible for more patients than with usual care, and increasingly have to deal with technology and data. It is conceivable that these factors influence health professionals’ attitude towards their work and their well-being, as their professional environment is changing considerably. The aspect of digitalisation of the professional environment merits further research.

Conclusions

Virtual care using telemonitoring of patients that are currently admitted to the hospital can save money, provided it is deployed at sufficient scale, designed to minimise time spent by health professionals, or the costs of the technology are considerably reduced. Presently, in many European countries with fully or partly capitated budget systems, the financial situation of hospitals might even suffer when venturing into virtual care for postoperative and comparable categories of patients if the aforementioned aspects are not taken into account, as a result of higher costs and lower incomes.