Previous page Adjust text size

Brief analytical summaries or syntheses #17

Predicting social care costs. A feasibility study


This study explores whether statistical models can be used to predict a person’s future need for intensive social care. Aside from the predictive models developed, this work points to the important potential of linked health and social-care data to support policy analysis and to guide the planning and delivery of services.


The social care and healthcare costs of people with complex needs are set to rise steeply in the U.K. over the medium term, due to the ageing population and the growing number of people living with long-term medical conditions. It will become increasingly important to find ways of helping local councils and health services take earlier action to support people so they can remain independent and at home as long as possible. The ability to identify these people would be helpful so that they could be offered targeted, effective support and preventive care aimed at promoting independent living, which could delay or avoid some health care costs. Over the last few years, many NHS organisations in England have started using predictive tools to work out which individuals in a given population are at risk of unplanned admission to hospital. These models use historic patterns in population data to make predictions at the individual level. This study applies the model to anticipate needs for intensive social care.

Analysis and results

There is high-quality evidence from the literature that certain interventions, such as multi-dimensional geriatric assessment in the home, can prevent or delay admission to a care home. However, such programmes are expensive, so objective ways of determining risk at the individual level are needed for councils to invest in them efficiently.

This predictive model focuses on two measures:

  • Sensitivity, which is a measure of how good the model is at detecting people in the population who will truly experience the outcome of interest (for example, admission to a care home).
  • Positive predictive value (PPV), which is a measure of the probability that the people identified at high risk will indeed truly experience that outcome.
  • Predictive models use information about past healthcare use to identify needs and then predict future healthcare needs. To predict social care costs, information on both health and social care needs is combined to predict future social care use.


This study reveals that despite the fact that health and social services interact for millions of people, their information systems tend to be discrete and distinct. This research shows how it is possible to link routine data from health and social care information systems in a way that protects individuals’ identities. Predictive models have the potential to provide a better experience for service users and to offer more cost-effective care. This project has shown that it is possible to construct predictive models for social care but it remains to be seen how these models might fit into everyday working practice. The quality of data about individual healthcare use has improved considerably over the past decade. Now a next step is needed to ensure that information about social care services improves in the same way. This will require strategies to improve the coding, collection and sharing of data in ways that protect confidential information.

Implications and recommendations

The predictive accuracy of the models in this study is comparable with that of the models used by the NHS to predict hospital admissions. However, the practical use of this model is less clear. The authors suggest pilot projects and testing as case-finding tools for social care in the participating sites.

Access to data: The linkage of health and social care data continues to raise questions about information governance processes. In order to exploit the huge potential of linked data there needs to be a better understanding of what is and is not permissible. The authors suggest that a clear protocol needs to be agreed by the National Information Governance Board for Health and Social Care and the Information Centre, and that this should be widely disseminated. The authors also believe there would be value in developing an experimental dataset that includes pseudonymous social care datasets linked with health data from a number of sites to enable more detailed analyses.

Data linkage should be promoted for the commissioning of integrated health and social care organisations.


Predicting social care costs. A feasibility study

Gouvernement du Québec
© Gouvernement du Québec, 2017