Risk adjustment and predictive modelling use relationships in administrative health information from the past to determine the expected future health care needed by each individual in a population. Risk adjustment can make comparisons of health services more meaningful. In addition, risk prediction methods can be used to help target services. This project aims to demonstrate how the analysis of routine data can be used to improve the access, equity and efficiency of health services.

This video presents the highlights from the 2013 Predictive Risk Summit, which explored the application of predictive risk modelling and evaluations of community-based interventions.

Risk adjustment has been used for many decades in research studies but since the 1980s the approach has increasingly been used in managing health care systems. Risk adjustment tools are used widely in the United States and in parts of continental Europe to help determine health payments – either for fixing ‘capitated’ budgets or for deciding reimbursement rates for individual patients. Several proprietary risk adjustment systems from the US are now seeking to expand into European markets.

To date, the most widespread use of risk prediction in the English NHS has been in the use of ‘case finding’ predictive modelling tools, such as PARR and the Combined Model. In August 2011 the Department of Health announced that it would not be commissioning a national upgrade of these two models, but the Nuffield Trust have been exploring a range of models that could be used by the NHS in England. Scotland and Wales have their own predictive models called SPARRA and PRISM respectively.

Guidance on predictive risk modelling for commissioners

To help commissioners, the Nuffield Trust have been exploring a range of models that could potentially be used by the NHS in England.

In our research paper Choosing a model to predict hospital admission: an observational study of new variants of predictive models for case finding (BMJ Open August 2013) we present a new set of models to predict unplanned hospital admissions using prior hospital and GP data.

With these models we tried to answer questions about how predictive performance was affected when various restrictions were placed on the modelling data. Restrictions on the availability of data will be familiar to many of those wanting to build their own models, and so we hope that our findings will help commissioners make choices about predictive models being built in their areas.

A summary of the work can be found in a report written for the Nuffield Trust Predictive Risk Conference 2013, and further information on the new models is available through contacting our research team.

In November 2011 we published more general guidance that explores how clinical commissioners should choose a predictive risk model based on factors including the outcome to be predicted, the cost of the model and its associated software, the availability of data, the accuracy of the predictions, and the preventive intervention to be offered on the basis of predictions. This follows an article in the Health Service Journal in October 2011.

We are aiming to demonstrate how the analysis of routine data can be used to improve the access, equity and efficiency of health services.

Our research summary: Predictive risk and health care: an overview (March 2011) provides a useful overview of how risk adjustment techniques are currently being used in the NHS, considering the principal applications of risk adjustment, namely:

  • Case finding: identifying individual patients at risk of a particular outcome such as unplanned hospital admission in the next 12 months. We have recently been using the principles of risk adjustment to develop a case finding tool for social care.
  • Resource allocation: changes to the way health services are organised and delivered in the NHS in England are coming at an important time in the development of risk adjustment techniques. Work on a person-based resource allocation formula for the NHS in England has been used to set GP practice budgets and could play an important part in setting GP consortium budgets.
  • Performance management: using risk adjustments when making comparisons between areas or organisations. This could apply to comparative analyses of Ambulatory Care Sensitive Conditions.

The increasing ability to link large data sets at an individual level whilst protecting the confidentiality of patients means that the range of data used on these types of models and their applications look set to grow. For example, the Nuffield Trust have also demonstrated that it is possible to build models based on linked GP, hospital and social care information that predict which individuals in a population are at risk of starting intensive social care in the next 12 months.

As well as publishing new materials, such as our research papers and our guide for commissioners, in July 2013 we also held the fourth of an annual set of one-day conferences. Access further details about this conference, including blogs and a highlights video.

Project outputs


Email to a friend

Your message will be:

I thought you might be interested in this page on The Nuffield Trust website.