Cookie policy notice

By continuing to use this site you agree to our cookies policy below:
Since 26 May 2011, the law now states that cookies on websites can ony be used with your specific consent. Cookies allow us to ensure that you enjoy the best browsing experience.

This site is intended for health professionals only

At the heart of general practice since 1960

Managing at-risk patients in virtual wards

Virtual wards in the community can care for people deemed to be most at risk of hospital readmission using a new predictive model, explains Dr Geraint Lewis

Virtual wards in the community can care for people deemed to be most at risk of hospital readmission using a new predictive model, explains Dr Geraint Lewis

Croydon PCT has been piloting the practical use of the NHS-owned Combined Predictive Algorithm (combined model) since May 2006.

The combined model uses hospital and GP data to predict the risk of unplanned hospital admissions.

In Croydon, patients with the highest risk scores are admitted to one of two pilot ‘virtual wards', each of which has 100 ‘beds'. In the future it is hoped that GPs will commission virtual wards.

The concept

Virtual wards use the systems of a hospital ward to provide case management in the community. To be successful in the long-term, each virtual ward needs to be permanently linked to a group of practices in order to develop close working relationships.

Ward staff and structure

Each virtual ward has a capacity to care for 0.3 per cent of patients at highest risk of admissions. Croydon's population is 340,000, so the catchment for each ward is roughly 34,000 – about one ward for every 15 GPs. We can adjust the catchment areas to compensate for geographical pockets of high risk (smaller catchment areas where there are many high-risk patients, and vice versa).

A community matron leads the day-to-day clinical work of the ward, supported by other staff (see figure 2). The other key member of staff is the ward clerk, an administrator who collects and disseminates information between patients, their carers, GP practice staff, virtual ward staff and hospital staff.

The community matron is in daily telephone contact with the duty doctor at each constituent practice. There is an open invitation for GPs to attend ‘ward rounds'.

The virtual ward also has close working relationships with hospices, drug and alcohol services and voluntary sector agencies.


On admission, the community matron conducts an initial assessment at the patient's home. This record, and all further entries by ward staff, is entered into a shared set of electronic notes.

A summary from the GP computer system is pasted into these ward notes before the initial assessment, so as to provide background information and avoid unnecessary duplication of work. The GP practice is informed of all significant changes to the patient's management.

Ward rounds

Virtual ward staff hold a ward round each working day, discussing patients at different frequencies depending on their circumstances (see figure 3). Ward rounds are held in PCT offices and practice meeting rooms, and some staff participate by conference call.

The community matron can move patients between different intensity ‘beds' according to changes in their conditions.


Every night an email containing a list of each virtual ward's current patients is sent automatically to the ambulance trust, local hospitals, NHS Direct and GP out-of-hours co-operatives. This is uploaded on to these organisations' clinical computer systems.

Should a virtual ward patient present to those services then staff there will be alerted automatically to the patient's status. They can obtain up-to-date details from the ward clerk and also arrange early discharge back to the virtual ward.


When a patient has been cared uneventfully for several months, the virtual ward staff may feel the patient is ready to be discharged back to the GP practice. Ward staff also receive a prompt when the patient's name drops below the 100 people with highest predicted risk in that catchment area.

The ward sends a discharge summary to the practice and a discharge letter, using lay terminology, to the patient.

For the first two years following discharge, the GP practice will conduct quarterly – rather than annual – reviews.

Lessons from the pilot so far

Patient characteristics The high-risk patients typically have one or more physical illness plus some kind of mental ill-health.

From the literature we were expecting there to be lots of patients with COPD but in fact they are extremely varied. Alcohol is a big issue. We were expecting patients mainly to be elderly, but they range from early 20s to the 90s.

An advantage of the combined model is that it allows you to make risk comparisons on the same scale for a toddler and an octogenarian – the first time the NHS has been able to do that.

GP buy-in Initially it was quite difficult trying to explain the idea in abstract terms.

GPs were sceptical, partly because of funding concerns and also the idea of a computer picking which patients should be admitted.

Nobody has got ‘admitting rights' to the wards – it's who the combined model says are at highest risk who are admitted.

Eventually we agreed on a memorandum of understanding between the practices and the PCT.

GP attitudes were transformed once they started to see which named patients we were going to work with. GPs are delighted that there is a whole team dedicated to looking after these complex patients.

No patients have yet been discharged back to GP practices.

The arrangement where the matron either phones straight through to the duty doctor or makes an appointment to see the patient's usual doctor has worked very well.

A talking shop? A potential criticism of the scheme is that it looks like a talking shop. Why do you need all these multidisciplinary team members sitting down for an hour every day? My counter-argument is that perhaps this is not happening enough at the moment.

We're finding there are huge opportunities to get rid of duplication. For example, we found one patient who during the course of a single week had five routine full blood counts ordered and three routine chest X-rays by different teams – simply because the left hand didn't know what the right hand was up to.

It's that kind of wastage we're trying to cut through.

IT One of the big difficulties was the clinical IT system used by virtual ward staff. We've used the PCT's computer system and tried to adapt it, but it's less than ideal as it does not allow remote access and is not joined to GP practice systems.

The beauty of the combined model it is that it is owned by the NHS so it's free to use. But because every GP uses a slightly different system, you need quite advanced skills to set up the combined model and extract the data to feed into it.

Currently we're only refreshing the model every six months because it involves someone collecting the data manually from each practice.

But we recognise that it's incredibly important to minimise the delay between prediction and admission to the ward. So we're getting a system built that will allow us to download the necessary data automatically from GP practices, in order to run the combined model monthly.

Commissioning virtual wards

The main extra resource considerations for practices wanting to commission this model are two full-time members of staff – the community matron (which PCTs are obliged to appoint anyway) and the ward clerk.

All the other ward staff juggle the work with their other day jobs. These patients existed anyway and the virtual ward is often about reorganising the way staff work and removing duplication.

However, we don't yet have evidence of whether this works or not and as a result it would be unreasonable to expect GPs to commission it at this stage. We're hoping by gathering evidence in a robust way that GPs next year might feel that they're keen to commission it.


You'll often hear of community matrons in other parts of the country who say they've had an 80 per cent reduction in hospital admissions, which sounds fantastic. But if their patients were not selected using predictive risk modelling then we know from the evidence that even if they didn't do anything at all with these patients, their number of admissions drops off rapidly.

This is called ‘regression to the mean', and it occurs within about 18 months. So some of the 80 per cent reduction was going to happen anyway, and we don't know how much effect the community matrons had per se.

The good thing about our patients is that they're found in a predictive manner, so any benefits we do show will be true benefits and not regression to the mean.

We are applying for funding to run a multisite randomised control trial and in the meantime are doing a case control study.

We have also just had the go-ahead to open another eight wards to cover the rest of Croydon.

Dr Geraint Lewis was specialist registrar in public health at Croydon PCT and now works as a visiting fellow at the King's Fund.


The combined predictive model explained

• The model was developed by the King's Fund, Health Dialog and New York University.

• It builds on their previous predictive tool, Patients At Risk for Rehospitalisation (PARR).

• PARR uses inpatient, outpatient and A&E data, while the combined model also uses GP data for 560,000 patients supplied from two PCTs.

• The model segments all patients into four groups of risk of future hospital admission

• All patients' personal identifiable data is stripped out and their NHS number turned into ‘gobbledegook' using a pseudonymisation machine, before being entered into the combined model. A list

is then produced in order of risk, and the ‘gobbledegook' can only be decrypted with the permission of the patient's GP.

• The combined model helps identify potentially greater cost savings. The net saving for the top 250 at-risk patients, once the cost of an intervention to reduce admissions is taken into account, is £95,248, compared with £63,265 idenfied with the PARR tool

Rate this article 

Click to rate

  • 1 star out of 5
  • 2 stars out of 5
  • 3 stars out of 5
  • 4 stars out of 5
  • 5 stars out of 5

0 out of 5 stars

Have your say