Impact of COVID virtual wards on hospital activity

Overview

This chapter draws on a manuscript by Georghiou et al.122 published in Lancet eClinicalMedicine (2022): https://doi.org/10.1016/j.eclinm.2022.101441. This is an open access article under the term of the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits use, distribution and reproduction in any medium, provided the original work is properly cited (see https://creativecommons.org/licenses/by/4.0/).

What was already known

• Post-hospital discharge virtual wards have been found to have a positive impact on patient outcomes when focused on patients with specific diseases, for example those with heart disease. There has been less evidence of impact for more heterogenous groups of patients.
• There has been little evidence thus far that post-hospital discharge virtual wards (using pulse oximetry monitoring) have helped to reduce the LOS of patients hospitalised with COVID-19 or to change rates of subsequent readmissions for COVID-19.

What this chapter adds

• This national-scale study has found no evidence that the roll-out of post-hospital discharge virtual ward services for patients with COVID-19 in England reduced lengths of stay in hospital, or impacted rates of readmission.
• While there is currently an absence of evidence of positive impacts for patients with COVID-19 discharged to a virtual ward, our study emphasises the need for quality data to be collected as part of future service implementation.

Introduction

There has been a long-standing ambition in the NHS in England to develop remote monitoring services to better support people with long-term conditions out of hospital.18 Virtual ward models have been proposed as one way of easing the transition of patients being discharged home, and there is some evidence that such models (distinguished from less intensive telemonitoring) may help to reduce mortality and readmissions, especially where the intervention is focused on patients with a specific disease, for example heart failure.123

In facilitating the national roll-out of CVW services in January 2021, NHSEI stated that ‘CVWs have been proven to reduce admissions/bed occupancy and improve LOS’, and so could act to mitigate the pressure on hospital beds, which was severe at the time.6 However, evidence of the impact of such services on hospital activity has been lacking. A US study found a reduction in rates of emergency department presentation or readmission for a group of 225 patients enrolled on to a post-discharge monitoring programme,124 and two studies (from the Netherlands, and a case study from England) estimated reduced lengths of hospital stays, but with small patient cohorts and uncontrolled methods.125,126 Elsewhere, where outcomes have been reported, they have tended to focus on rates of patient escalation (including readmissions) in the absence of a comparator. Occasionally, outcomes are considered against a perceived optimal level (e.g. a 10% conversion rate to admissions for ambulatory care pathways),31 or the reported performance of services elsewhere.25,31,36,121,127–129

Data on which patients were enrolled to CVWs in England were not collected, but information was assembled on launch dates of CVWs at each hospital trust. By late February 2021, the large majority of acute hospital trusts in England had a service available for use. The analysis described in this chapter aimed to use patient-level admissions data to provide quantitative evidence of the impact of the service, to contribute to future development of post-discharge home monitoring programmes during the remainder of the COVID-19 pandemic, and beyond.

The primary aim of the analysis described in this chapter was to address the following research question:

What was the impact of CVWs on the length of COVID-19 hospitalisations and rates of subsequent readmission for COVID-19?

A secondary aim, in the context of COVID-19s widely reported differential effects on outcomes for groups of patients (e.g. by gender, age and ethnicity),12,130 was to reflect on associations found between patient characteristics and hospital outcomes.

Methods

A summary of the methods is provided below; further details can be found in Chapter 2 and Report Supplementary Material 1.

We used HES APC data to extract information on patients with a confirmed or suspected primary diagnosis of COVID-19 (ICD-10 codes U071 or U072) who were discharged alive between 17 August 2020 and 28 February 2021 (during England’s second COVID-19 wave). We limited our analysis to 123 acute hospital trusts for which there was a known CVW service start date (from a data set collected by KSS AHSN).

Using patient-level (pseudonymised) data, we developed multivariate models to examine the impact of the availability of CVW (the primary independent variable of interest) on two outcomes: the length of inpatient stay for COVID-19, and on subsequent readmissions for COVID-19. Factors included in our models are shown in Table 8.

For our main analyses negative binomial regression was used to examine the relationship between independent variables and LOS, and logistic regression was similarly used for occurrence of any readmission. To account for clustering at the level of the hospital trust, generalised estimating equation approaches were used. Descriptive statistics and a number of sensitivity analyses were also carried out.

Results

Participants

We found 139,619 discharges between 17 August 2020 and 28 February 2021 where the patient had a primary diagnosis of COVID-19 recorded during their inpatient stay, and had not died in hospital. These discharges belonged to 129,461 individuals. Table 9 shows characteristics of these COVID-19 patient discharges.

The number of patients with COVID-19 discharged in this time across England (not limited to the 123 analysis trusts) was 142,216. Thus, our analysis includes 98.2% of recorded inpatient stays of patients in England with a primary COVID-19 diagnosis, discharged alive.

Table 9 also shows the characteristics of the patients split into two key analysis groups: patients discharged where no CVW was available (N = 81,174, 58.1%), and those discharged where a CVW was available (N = 58,445, 41.9%). These two sets of patients were well matched in terms of gender and comorbidities. Patients discharged where a CVW was available were more likely to be in the age range 50 to 64 and members of non-white ethnic groups, and less likely to be from the most deprived areas. As a consequence of the accelerated set-up of CVW services from early January 2021 (Figure 20), discharge dates of patients where a CVW was available tended to take place towards the end of the analysis period, with nearly four in five occurring on or after 4 January 2021 (compared with two in five of the discharges where a CVW was not available). Discharges of patients taking place where a CVW was available were more likely to be occurring in the context of higher levels of beds occupied by patients with COVID-19.

FIGURE 20 Cumulative start dates of CVW services in England’s non-specialist acute hospital trusts (n = 123 trusts); 14 trusts had no CVW by 22 February 2021. The grey box represents the main period of this analysis. (Figure not reproduced here; see the original article.)

COVID-19 length of stay

The mean LOS for patients discharged where a CVW was available was 10.1 days [standard deviation (SD) 11.3 days], compared with 8.9 days (standard deviation 10.3 days) for patients where a CVW was not available (Figure 21). This was an unadjusted incidence rate ratio (IRR) of 1.13, or 13% longer for discharges taking place from hospital trusts with a CVW. Distributions of LOS for COVID-19 for patients discharged with and without an available CVW are provided in Report Supplementary Material 5.

FIGURE 21 Mean (unadjusted) length of COVID-19 hospitalisations and adjusted IRRs, with respect to categories of factors in multivariate model. (Figure not reproduced here; see the original article.)

Fully adjusting for all covariates, we found an adjusted IRR of 1.05 (95% CI 1.01 to 1.09) for discharges where a CVW was available compared with those where a CVW was not available, indicating that the LOS was 5% longer for relevant COVID-19 inpatient stays where a CVW had been available.

We additionally found strong, generally positive gradients of LOS for increasing age and comorbidity (note that these relationships relate to all 139,619 discharges and not just those with a CVW available). Females had shorter lengths of stay than males (adjusted IRR 0.95, 95% CI 0.94 to 0.96), as did all non-white ethnic groupings, compared with the white group (e.g. Asian patients had an adjusted IRR of 0.89, 95% CI 0.87 to 0.91, and black patients an adjusted IRR of 0.93, 95% CI 0.90 to 0.95). Patients resident in the most affluent population quintiles tended to have slightly shorter lengths of stay than those in the most deprived quintile. First COVID-19 inpatient stays were longer than subsequent stays (adjusted IRR 1.07, 95% CI 1.02 to 1.12). The LOS shortened significantly for all time periods after 31 August 2020, until mid-February 2021. As the proportion of acute beds occupied by patients with COVID-19 increased, LOS tended to fall: for every ten-percentage-point increase in this proportion, the adjusted IRR was 0.97 (CI 0.96 to 0.98).

Model variants carried out as sensitivity analyses marginally altered the resulting adjusted IRR of the CVW-available discharges (see Report Supplementary Material 5). Shortening the time period variable to seven days reduced the IRR slightly such that the difference was no longer significant at 95% statistical confidence level (IRR 1.04, 95% CI 0.997 to 1.08), while increasing the period to 28 days increased the adjusted IRR to 1.08 (95% CI 1.04 to 1.13). The addition of earlier (wave 1) data made the two groups effectively indistinguishable from one another (IRR 1.01, 95% CI 0.97 to 1.05). The two alternative LOS outcomes tested had little impact on the adjusted IRRs compared with the base model.

COVID-19 readmission within 28 days

Of the patients discharged from trusts with a CVW available, 13.0% were readmitted with COVID-19 within 28 days, compared with 13.2% from trusts where no CVW was available (Figure 22). Adjusting for all variables, we found a non-significant difference in COVID-19 readmissions within 28 days for discharges where a CVW was available compared to those where a CVW was not available (adjusted OR of 0.97, 95% CI 0.91 to 1.03). Our sensitivity analyses demonstrated the robustness of these findings (see Report Supplementary Material 5).

FIGURE 22 Numbers (%) of patients readmitted for COVID-19 within 28 days and adjusted OR with respect to factors in multivariate model. (Figure not reproduced here; see the original article.)

As with LOS, we found generally positive gradients of relative odds of readmission with increasing age and comorbidity. Females had significantly lower odds of readmission than males (adjusted OR 0.78, 95% CI 0.75 to 0.81). Black patients had lower odds of readmission compared with white patients (adjusted OR 0.91, 95% CI 0.84 to 0.99), while no other ethnic categories (for patients with known ethnicity) showed similarly significant differences. There were no apparent differences in the odds of COVID-19 readmission for patients resident in less deprived areas compared with those in the most deprived quintile areas.

For both LOS and readmissions models, adding variables one by one to our multivariate models (an example is shown in Report Supplementary Material 5) revealed that the addition of the 14-day time period categories, and the clustering by hospital trust had large impacts on the estimated adjusted ratios, but all additional variables (age, comorbidity index, and so on) had relatively marginal additional impacts.

Discussion

Key findings

In this analysis of patients with COVID-19 discharged from hospital during the second wave of the COVID-19 pandemic in England, we found no evidence of a relationship between the availability of hospital-based CVWs and subsequent rates of readmission for COVID-19. We did, however, find that the LOS for patients hospitalised with COVID-19 was 5% longer where a CVW was available.

Comparisons with other studies

There are few studies available against which we can compare our results. Our finding of no difference with respect to readmissions appears to contrast with that of an analysis of a post-discharge remote monitoring model implemented in five Boston (MA) hospitals. This study found that enrolled patients had a reduced odds (OR 0.54, p-value 0.04) of attending emergency department or being readmitted within 30 days of discharge, compared to those not enrolled.124 However, given the relatively large CIs in their estimate, emergency department attendances being much more common than readmissions for the enrolled patient cohort, and those readmissions being for any cause, our findings may not be inconsistent.

Two other analyses have claimed reductions in length of hospitalisation associated with post-discharge models, on a scale of 30–40%. One, related to implementation in a Netherlands hospital, estimated a reduction in the LOS of 5.0 days (with a resulting mean LOS of 10.6 days),125 and a second, in an English hospital trust, reported a reduced LOS of 10 days, against an average of 17 days prior to implementation.126 Both analyses, however, were based on small patient cohorts with no formal statistical testing. In the first analysis it was not clear how estimates of reductions were made, and in the second there was no attempt made to control for differences between the compared groups.

More generally, there is evidence of non-COVID-19-related post-discharge virtual wards having a positive impact on readmissions when employed as a disease-specific intervention (at least in the case of heart failure), but there is not similar evidence of impact for more mixed groups of patients at high-risk of chronic disease.123 Our findings may be consistent with this picture, considering the heterogeneity of patients with COVID-19, and the multiorgan effects of COVID-19 infection.131–133

Strengths and limitations

With respect to the analysis of post-discharge COVID monitoring services, our study is currently unique in its scope. We have made use of a national administrative hospital data set, and have analysed almost 140,000 COVID-19 admissions in an effort to detect an impact of a national-scale roll-out of post-discharge remote monitoring services. We have been pragmatic in our use of the available data and have controlled for characteristics available to us.

Nevertheless, there are several limitations, and our results should be interpreted with caution. We did not know which patients were enrolled on to a CVW, and so treated all patients as potentially having received the intervention where one was available at a trust. We had no information about important clinical factors (e.g. ICU admission, clinical readings, including oxygen saturation levels, and specific treatments received). The location to which patients were discharged (home, care home, etc.) was not known, and we also had no information about out-of-hospital deaths (and this during a period in which rates of out-of-hospital deaths had been persistently above long-term norms).134 We extracted data only on patients discharged alive, so the impact (if any) of patients with COVID-19 who died during their hospital stay has not been accounted for. Our analyses have assumed that these factors, which are likely to have been associated to a greater or lesser degree with eligibility for referral, and/or with risks of readmission and long stays, were well balanced between the two key groupings of patients after having adjusted for model factors, but the extent to which this is the case is not known. However, in terms of at least one of these missing factors, an analysis of a cohort of 419 patients with COVID-19 managed on a post-discharge pathway suggests that out-of-hospital deaths may have been relatively rare (<0.1%).120

As outlined in Chapter 4, for seven hospital-based sites we were able to estimate a range of 4–65% of discharged patients with COVID-19 may have been monitored on a CVW. But nationally, we did not know the scale of each trust’s implementation and our analysis wasn’t able to examine potential dose–response relationships.

We also know that services were implemented with significant differences, depending on the locally adopted models of care. Our analysis treated the intervention as being homogeneous, and was not able to estimate differential impacts by hospital trust, although these impacts may have existed.

Interpretation

We found longer COVID-19 inpatient stays associated with implementation of CVWs; this was not expected for a service that aimed to support the early discharge of patients. One explanation might be that CVW eligibility criteria influenced hospitals’ discharge decisions such that patients were kept in a bed for longer than they otherwise might have been. It is not clear whether local pre-hospital (CO@h) services starting at similar times might have affected admissions in such a way that patients were admitted sooner and stayed for longer. The finding might, in part, also be a consequence of many CVW services launching during the falling (improving) edge of England’s second COVID-19 wave, with longer lengths of stay being a consequence of improving bed capacity.

We did attempt to control for the latter effect in our addition of trust-specific weekly occupied beds data, and with our inclusion of a time period variable in our multivariate analyses. In one sensitivity analysis, when we shortened the time period to seven days (from 14), we no longer found a statistically significant difference in LOS at the 95% level (adjusted IRR 1.04, 95% CI 1.00 to 1.08, p-value 0.07). The time factor used in our fully adjusted models was somewhat arbitrary, and it is not clear that 14 days was better used above seven days. The finding of a longer LOS associated with CVW therefore needs to be considered with some care.

The expected impact of such services on readmissions following discharge is not obvious. These services aimed to detect early deterioration for more rapid escalation to potentially improve ultimate outcomes. It may be the case therefore, that readmissions might have been expected to increase. Unfortunately, with limitations in the available data described above, we were not able to assess any changes in the severity of patients on readmission.

As a result of the limitations in the available data outlined above, there is a strong possibility that any positive impacts – at a local hospital trust level, or for specific versions of the pathway – were not ultimately able to be detected. This limits the extent to which this analysis can contribute to the further development of these, or similar services, and highlights the importance of making routine data collection – appropriate for use in evaluation – an intrinsic part of services’ operations.

Briefly, we reflect on observed relationships between patient characteristics and COVID-19 lengths of stay, and readmissions. Positive relationships between age and comorbidities and these outcomes have been observed elsewhere.135–138 Meanwhile, differences with respect to gender (in this study’s case both outcomes were lower for females vs. males) tend to have been absent.131,136,138,139 We found shorter lengths of stay and lower rates of readmissions for patients belonging to specific ethnic groups with respect to white patients (this was the case for patients belonging to black ethnic groups for both outcomes, and also Asian, mixed and ‘other’ ethnic groups for LOS). Evidence from elsewhere is mixed, but at least one US study found shorter lengths of stay for hospitalised non-white patient groups.138 Ultimately, however, care is needed when interpreting findings for specific characteristics separate from the context of the other model variables. It is possible, for example, that non-white ethnic groups (to give one example) may be acting as fine-scale markers of urban or more deprived areas (in addition to the model deprivation categories) and as markers of younger patients (within modelled age bands).

Conclusion

Our analysis has not shown any evidence of early discharges or changes to rates of readmissions associated with the roll-out of CVWs in England. While this may reflect the true impact of the service, it may be, in part, a consequence of the lack of certain data: on which patients were enrolled (or even how many), what COVID-19 treatments were received while in hospital, and their clinical observations. It is possible that CVW services had a range of impacts that we have ultimately not been able to reveal.