Research suggests economic model for virtual wards not viable on hospital activity alone

image of graph

This new report is an important addition to the evidence base specifically on case management and more generally in terms of interventions to reduce hospital admissions but, consistent with other studies in this area, can’t give definitive answers.  Much of what already exists in relation to virtual wards is anecdotal or lacks detail and it’s often difficult to apply or generalise learning to local contexts.

Many commissioners have set priorities relating to unplanned admissions particularly in relation to long term conditions and will be interested in the key messages from this research.


  • The researchers explored the impact of virtual wards on rates of A&E attendance, social care provision, GP practice visits and use of community health services, across 3 sites (Croydon, Devon and Wandsworth).  The researchers collated data relating to 989 patients attending Virtual Wards during a period ranging from 2006 to 2010; patients were followed up for 6 months and compared to controls identified through local and national data.
  • hospital bed

    The median length of stay was 338 days; 25% stayed 144 days or less and 25% stayed 720 days or more.

    Overall, the researchers found no evidence of reductions in emergency admissions, ambulatory care sensitive admissions or hospital costs.  Emergency admissions reduced by 0.27 per person per 6 months, compared to 0.6 in the control group, suggesting a net difference increase of 0.33 per person per 6 months (p<0.01). Visits to A&E increased by 0.38 per person per 6 months relative to controls.

  • The researchers did observe a reduction in elective admissions and outpatient attendances and suggest this could be due to greater co-ordination of care.  There was a net reduction of 0.81 outpatient attendances per person per 6 months against controls.  The outpatient speciality showing the greatest reduction was haematology – which could be due to increased monitoring of INR results in community.  There was no clear explanation for the reduction in elective admissions and the impact did not appear stronger in any particular specialty.
  • GP visits increased by 1.57 per person per 6 months but reduced by 1.29 in controls, suggesting a net difference increase of 2.86 visits per person per 6 months.  Contacts with community nurses increased by 8.67 per person per 6 months relative to controls.

Cost analysis

  • The research also included a cost analysis, exploring the costs of setting up and running virtual wards.  The study shows a large variation in costs (even allowing for expected local variations from differences in rent, labour market etc) – from £3 per patient day in Croydon/Devon to £17 per patient day in Wandsworth; this is to an extent explained by the greater involvement of GPs in the Wandsworth model.   Over the 6 month follow up period, the cost per patient varied between £510 to £2890.
  • The researchers state: “In order to have broken even, the virtual wards would have needed to achieve a greater redcuction in emergency admission reates in the first 6 months after starting the intervention.  For the virtual wards with lower running costs, this could have been around a 10% fall.  In contrast, for the more costly models, the fall in emergency activity would have needed to be much greater – up to 100%.”

Limitations of the research

There are some important limitations to the research, acknowledged by the authors:

  • Outcomes are compared with matched controls using both local and national data.  The comparison with nationally matched controls cannot rule out other interventions which may be in place to reduce admissions therefore the researchers are cautious about interpreting neutral findings.
  • There was considerable variation across the 3 sites in the definition of “virtual ward” and how the service was delivered, for example, the involvement of GPs in Wandsworth and the move in Croydon away from a multidisciplinary model.  The study was not powered to evaluate effects of different models.
  • The sample was dominated by cases from Croydon which makes it difficult to draw out lessons from Devon and Wandsworth as the number of cases were not statistically significant.
  • The 6 month follow  up period may not be sufficiently long to demonstrate impact.
  • The costs do not include prescribing costs due to limited data access (costs included staffing, travel, land, computers, management).
  • The researchers focused solely on outcomes relating to service utilisation – the research does not explore other outcomes such as patient experience, improvement in patient outcomes, quality of life.

Key learning for commissioners

Learning from this research suggests the following critical success factors to virtual ward or similar services:

  • Leadership and engagement to maintain momentum and to avoid straying from the model agreed, during transition from pilot to business-as-usual.
  • A multidisciplinary model ensures broader ownership and engagement; the process for access to specialist geriatrician support also needs to be clear.
  • Clear criteria are required to manage selection and access and ensure a focus on higher risk patients.
  • The timeliness of the intervention is critical – the researchers note a time lag between identifying patient at risk and enrolment onto the virtual ward.
The use of robust predictive risk tools can help focus on higher risk patients and avoid the regression to the mean effect where outcomes may have improved regardless of the intervention.

The use of robust predictive risk tools can help focus on higher risk patients and avoid the regression to the mean effect where outcomes may have improved regardless of the intervention.

The sites researched each had a different approach to selection of patients and their reliance on predictive risk tool; Wandsworth include referral by GPs which may have helped increase a sense of ownership within primary care.

The volume and complexity of case load needs consideration.  The researchers quote Department of Health guidance which suggests a volume of 50-80 patients and cites analysis that higher case loads are associated with more reactive care and increased admissions; however, smaller caseloads may increase quality but may not be as cost effective when opportunity costs are taken into account.

There were considerable variations across the 3 sites studies in terms of level of staffing, types of staff used, breadth of responsibilities for staff and length of stay.

The report includes reference to recommendations for practice from a related article by Lewis et al (, including:

  • use of predictive modelling;
  • use of impactibility model to identify high risk patients amenable for preventive care
  • organisation of virtual wards around groups of GP practices;
  • catchment areas reflecting distribution of high risk patients to ensure equity;
  • staff mix to reflect needs of local high risk patients;
  • appointment of a  ward clerk;
  • clear out of hours arrangements;
  • a single patient record;
  • alert systems to notify if patients use urgent care services;
  • regular ward rounds;
  • a clear discharge policy;
  • staff training in admission avoidance techniques e.g motivational interviewing;
  • take account of Roemer’s Law, which suggests that if a bed is freed up, it may still be used]
  • continuous monitoring and feedback.

Detailed descriptions of the 3 models are included as an appendix, outlining staffing arrangements, target population, caseload, discharge etc.


Lewis G et al (2013), Impact of “Virtual Wards” on hospital use: a research study using propensity matched controls and a cost analysis.  NIHR.  Available at

Related documents

Lewis G et al (2012), Multidisciplinary case management for patients at high risk of hospitalization: comparison of virtual ward models in the United Kingdom, United States, and Canada, Popul Health Manag. 2012 Oct;15(5):315-21.  Abstract available at

The Kings Fund has recently published a report, Proactive case management using the community virtual ward and the Devon Predictive Model, available at

Share on Facebook Tweet this on Twitter Share on LinkedIn Share on Google+