Mathematical modelling can help predict risk of hospital admission in a given population, writes Daloni Carlisle

Some ideas are flash in the pan, lasting until the next thing comes along. Others stick around, slowly developing and spreading. Predictive risk modelling is one of the stickers.

At its most basic, predictive risk modelling uses mathematical models to make predictions about events such as which people in a given population are the most likely to be admitted to hospital in the next 12 months. It has been around for several years, coming to the fore in the mid-noughties when the NHS was faced with reducing emergency admissions by 5 per cent by 2008.

Geraint Lewis was a public health doctor in Croydon at the time and, with his community nursing colleagues, he reasoned that if he could use routinely available data to predict which patients were most likely to need re-admission in the next 12 months, nurses could target interventions to prevent them. The virtual ward was born.

This combination of maths and matrons proved fruitful. The Department of Health commissioned the King’s Fund, New York University and Health Dialog to develop two predictive tools. The first was the Patients at Risk of Readmission case finding tool (PARR), which uses inpatient data to predict the likelihood of readmission.

In 2006, the group developed the combined predictive model, which uses data from inpatient episodes, outpatients, accident and emergency and general practice to predict which people in a population are most likely to need emergency admission to hospital the following year.

Research programme

Fast forward to 2010 and the use of predictive modelling has spread across the UK and virtual wards have gone international (see box).

Dr Lewis is now senior fellow at the Nuffield Trust, which has a large research programme on predictive modelling and evaluating whether interventions such as virtual wards reduce emergency hospital admissions in practice.

Meanwhile, Health Dialog has been taken over by Bupa and is approved under the framework for providing external support for commissioners to provide commissioning support to primary care trusts.

Bupa associate director of analytics at Health Dialog Ian Manovel says: “We estimate that 80 per cent of PCTs are now using predictive modelling.”

He has worked with a number of them to develop tools that help GPs and PCTs generate and manage their “lists” of patients at risk of hospital admission. For example, Health Solutions Wales (the Welsh equivalent of England’s NHS Information Centre) released a web-based all-Wales tool in April 2010. NHS West Midlands has developed an integrated care manager tool for all its 17 PCTs; Commissioning Support for London is looking to do something similar across 31 PCTs.

The way virtual wards are run has developed, too, and Dr Lewis says there are now at least four approaches:

  • nurse led, with GPs becoming involved with individual patients as required (for example, NHS Croydon);
  • employing dedicated “virtual ward doctors” (for example, NHS Wandsworth);
  • general practice-led (for example, NHS Devon, see case study);
  • virtual discharge ward, running the predictive model on inpatients and offering those at high risk of readmission 30 days of support (for example Toronto).

But while each of these has been evaluated locally, the hard evidence for them reducing hospitalisation rates is scant.

Dr Lewis says: “We know they are intuitive and patients seem to love them, but we do not know whether they actually work.”

Predictive risk modelling is now being extended to predict other events. The Nuffield Trust has been commissioned by the Department of Health to conduct a feasibility study of building predictive models for social care. The aim of such models would be to use routinely collected health and social services data to forecast which people in a population are at greatest risk of incurring social care costs through loss of independence due to ageing and ill health.

Dr Lewis says: “Emergency admission lent itself to predictive modelling because it is undesirable, expensive, recorded in routine data and potentially avoidable. I would argue that admission to a nursing or residential home meets the same criteria. Similarly, admission to an intensive care unit or an extended length of stay.” l

Up and running: the Devon rollout

NHS Devon is about to roll out virtual wards across the county following a successful GP-led pilot.

GP Paul Lovell, from South Molton Healthcare Centre in north Devon, says: “We started our virtual ward in 2008, using existing community matrons and hiring an administrator to act as ward clerk.”

Now it is being rolled out to a population of 750,000, with 170 GP practices.

GPs will be using the combined predictive model and NHS Devon is hoping to commission a “front end” - a web-based system, complete with clinical dashboards, for accessing and managing lists of patients.

The virtual wards will be staffed by the existing complex care teams and GP practices working together, says Dr Lovell.

He adds: “What we found with the pilot was that the lists include a wide range of people, for example your mental health patients. You have to look at the skills in the team and the list and decide who is suitable for case management and what skill mix you need in your teams.”

Action points

  • Join the VW discussion group on yahoo
  • Decide which risk model to use
  • There are plenty of commercial providers offering help and support with predictive modelling. However, many of the most widely used tools are free to health service users
  • Decide how to manage cases and what interventions to offer

Virtual wards are up and running or in planning in the following PCTs:

1 Avon Valley, Hampshire

2 Croydon

3 Hillingdon (in planning)

4 Camden

5 Lambeth

6 Lewisham (in planning)

7 Wandsworth

8 Bedforshire

9 Norfolk

10 West Kent

11 West Sussex (in planning)

12 Surrey (in planning)

13 Somerset (in planning)

14 Manchester

15 Western Cheshire

16 Kirklees

17 Devon (planned, county wide)

18 Carmarthenshire

19 Powys

20 Grampian

21 Tayside

22 Essex

23 Bedford

24 Tibshelf, Derbyshire

25 Milton Keynes

26 Oxford

27 Stoke on Trent

28 Warwickshire


British Columbia






(Sources: Nuffield Trust; Bupa Health Dialog)