This site is intended for health professionals only


How we set up virtual wards



The model

In 2007, I attended a presentation by Dr Geraint Lewis, now a senior fellow at the Nuffield Trust, outlining the virtual ward concept and the principles of predictive modelling that he had piloted in Croydon, South London.

While the name ‘virtual ward’ has appeared in several areas since, including in secondary care, the original principle remains the combined use of two components – a predictive model to identify those at risk of admission and the use of a community-based, multidisciplinary team (MDT) providing secondary-level care at home to prevent admissions.

This seemed like an ideal model to adapt to our area, to help address the rise in patients with complex needs who were frequently admitted. We found to our frustration when we reviewed cases it was often clear that a co-ordinated intervention several months earlier might have prevented many of the admissions for that individual.

What we did

In 2008, we got the agreement of two other local practices and set out a business case to obtain a small amount of funding. We then set up a local ‘virtual ward’ in North Devon. This coincided with the development of 23 complex care teams (CCTs) across Devon, as part of a Department of Health-funded project.

Our core community services were developed into a multidisciplinary, multi-provider model by restructuring the existing workforce and employing a co-ordinator for each CCT. The teams scheduled weekly MDT meetings that included social workers, volunteers and community nurses.

Their remit was to provide co-ordinated community care to a local population. The inclusion of extra community matrons also increased capacity.

The virtual ward was a formal structure for the CCT’s case-management function to work within, and it helped identify and ‘admit’ individual patients. Practices received copies of the at-risk patient list and discussed these patients with the CCT to agree who should be admitted to the ward. Contact was made with the patient, consent gained, information gathered and goals for the period of admission set. This then led on to timely discharge from the service on achievement of those goals and space for a new admission.

It therefore provided additional value over and above establishing good communication within an MDT team (including primary care) by adding the drive of targeted case management and a process to glue it together and resist organisational drift back to traditional approaches.

Our principles of the actual case management are in accordance with a review article on the subject by the King’s Fund, published in 2011. The pilot project in our three practices and local CCT started in October 2008. We and other areas had used the PARR score to identify those at risk of hospital admission.

We evolved a new way of risk-scoring patients, based on the Patients At Risk of Readmission (PARR) score but including patient data as in the Combined Predictive Model (CPM). Local analysts published a variation of the system known as the Devon Predictive Model (DPM) – scores from this model are now shared with practices monthly using a secure website.

The accuracy of using the PARR score to identify patients improved significantly with the advent of the Combined Predictive Model (CPM), which also included the available data held in primary care on each patient. In Devon, we chose to implement the CPM in 2008 at the same time as running the virtual ward pilot. Commissioners and the local IT team selected and refined the variables and data validation in the existing system. We also developed a data-extraction service including information from hosted GP systems.

In 2011, a senior analyst on the project completed a local recalibration of the CPM, further increasing its accuracy. From this point, it was known as the Devon Predictive Model (DPM). We produce DPM scores for the local population on a monthly basis and share them with each practice via the secure Commissioning Informatics website.

The pilots appeared to have a clear impact on outcomes, with reduced visit requests, consultations and admissions for the high-risk group.

In 2010, the need to tackle a rise in admissions across the whole of Devon was significant enough that NHS Devon made the decision to roll out both the DPM and virtual wards to all practices and CCTs in the area.

NHS Devon incentivised this rollout through a LES. This required practices to complete a data-sharing agreement, establish monthly extracts, review the DPM lists and refer patients with the highest risk scores whose situation was deemed amenable to intervention. Practices were paid £120 per virtual ward bed per annum.

Subsequent versions of the LES have evolved to be more specific and focus on increasing the rate of ward occupancy, rate of patients admitted from the highest-risk DPM score group and supplying an out-of-hours care plan for each patient managed on the virtual wards.

In 2011, NHS Devon was divided into three localities, all of which had a slightly different approach to running the virtual wards. Although this affected the ability to ensure standardisation, it also provided us with some insights into what makes the intervention work.

Challenges

When our pilot started, the other 22 CCTs across Devon were providing a standard case-management function through direct referrals or the identification of individuals by the team that they thought may benefit from intervention. This approach was evaluated by the Nuffield Trust, which demonstrated there had been minimal impact on admissions except for a small group. When they ran the CPM retrospectively, the small group that received the benefit appeared to have high scores. This outcome suggested that the targeting of this case-management intervention and resource was essential to the success of complex care interventions.

While we were initially using the CPM to identify patients, we also asked GPs to nominate patients they thought would benefit from a ‘proactive approach’. Once both lists were ready, we compared them and found significant differences. There was some overlap, but the ‘clinician knows best’ default was challenged by the listings and some doctors were encouraged to take a leap of faith as we relied on the risk stratification tool when we rolled the service out.

During the roll-out it became clear we could only standardise to a point. The important parts were to ensure the right people were being seen, the turnover was occurring and the process flow happening. We had to adapt the wards on a team-by-team basis – for example, noting the differences between rural (moorland and coastal) and urban areas when we rolled the programme out across the county. This included challenges such as covering large areas, providing accommodation for the CCT members and introducing more mobile solutions into the IT system.

Results

Measuring the impact of admission avoidance schemes is difficult and open to many confounding factors.

The Nuffield Trust has developed propensity matching, a form of case-control that matches a local individual with another from a different, but similar, area in the UK. We haven’t been able to use this option because of the scale and skills we’d need to run it in Devon. We’ve also been hampered by the lack of any local controls, as the need for a result necessitated making the service available to any individual who was identified as at high risk of admission.

We’ve been looking at the effects on the target population – namely, the top 0.5% stratified by DPM risk score. We have then observed each monthly cohort over time, comparing rolling growth and outcomes to the cohorts of the corresponding months in previous years. Since the inception of the virtual wards just over two years ago, we have case managed nearly 6,000 individuals from these cohorts across Devon – approximately 45% of the available capacity.

The Devon population is over 760,000 and therefore the 0.5% population represents nearly 4,000 individuals at any one time. Within the 0.5% population, therefore, not all those patients would have been case managed over a year, due to appropriateness of the intervention and the scale involved, as well as the fact that each locality had different activity levels.

In terms of the outcomes seen, it’s important to note that there were no other specific proactive admission avoidance interventions targeting this group over the same period.

The results show an impact proportional to the intervention provided, which suggests a positive link. There appears to be a strong correlation between the responsiveness of the service to rising risk scores and duration of stay, with an optimum duration being around five or six months. If we achieved this length of stay more often, the improved turnover would also increase the capacity of the service further.

The results from the three localities also show the opportunity for improvement in each area, with some practices currently not hitting the full targets of the LES in terms of occupancy and profile and a few not participating. This reflects the difficulties of making changes on such a large scale and the challenges in changing attitudes, perceptions and working practices.

Locality A (the South) has demonstrated the greatest impact, with a 22% reduction in admissions and a minimum payment by results saving of £845,310. From comparing the different profiles of the service, it became clear that the role of the commissioner was key – they were the ones to actively support the project on the ground and troubleshoot. It was also clear, after the removal of the LES by one of the localities for a period, that this had a significant impact on the success of the intervention.

Having local GP champions in each cluster or in every practice also helped to bridge the gap between the theory and effective local implementation.

Ideally, as in the original Croydon model, a patient’s GP should be available to discuss their care at the weekly MDT meetings.
But we’ve found that to be impractical in terms of cost, time and geography. Communication between clinicians is vital to keep the process moving effectively, with monthly bed-state meetings with practices to review the lists plus weekly and ad hoc contact with case managers for clinical discussions encouraged. 

We are actively looking at technology and the use of video conferencing to facilitate this in conjunction with improvements in the local IT infrastructure and further integration and interoperability of IT systems. The use of our urgent case dashboard is changing to allow automated alerts and a centrally hosted overview of the virtual wards, as well as status updates for each area. From our experience of the factors involved in ensuring success, we will also be providing practices and CCTs with more detailed reports on their performance, which will be benchmarked. The impact of the LES on the rollout was significant, as prior to it only 12 practices had signed up for the DPM, but after a year we were supplying 112 practices with lists. We now also provide support or listings for several other PCTs and CCGs, and have started piloting and developing other predictive algorithms such as those for specific conditions, such as COPD.

 Locality Average virtual ward (VW) occupancy Average  occupancy from top 0.5% risk group Average  occupancy from low risk group (where DPM score < 19)  Percentage of practices hitting 2011 LES targets Q1 net admission change 2009 to 2011 for top 0.5% cohort Change in actual PbR cost associated with net admission change
Locality A (Southern) 107% 45% 16% 66% -22% -£845,310
Locality B (Northern) 99% 38% 22% 43% -14% -£210,758

Locality C

(Eastern)

90.3% 28.7% 26.4% 19.2% 3.91% £151,742
 

The future

It has been chosen as one of five national demonstrator sites for the King’s Fund Care Co-ordination Project to assess how integrated working can impact on communities. The report will look at how virtual wards and DPM combine in the existing integrated service structure in Torbay and how this could be translated to other areas of the UK.

The challenge in Devon is to take the service to the next level by encouraging its full adoption. This will include an increased focus on identifying and proactively responding to patients identified as at high risk.

The evidence nationally and internationally for the success of virtual wards, predictive modelling and case management has to date not encouraged large-scale uptake. From our experiences, however, this approach can deliver the savings needed and change the emphasis of care to a proactive, community-based approach.

Dr Paul Lovell is a GP in South Molton, Devon, and an urgent care clinical lead for NHS Devon