Predicting future costs of social care

This project was commissioned by the Care Services Efficiency Delivery unit (CSED) at the Department of Health.  The brief was to explore the feasibility 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 individuals in a population are at greatest risk of incurring social care costs through loss of independence due to ageing and ill health.

There is high-quality evidence that admission to a care home can be delayed or prevented through preventive, ‘upstream’, interventions (Predicting who will need costly care).  Clearly, such upstream interventions will be most cost-effective when they are offered to people who would—without intervention—truly go on to require intensive social care.  Therefore, if more efficient investment is to be made in preventive interventions, councils need ways of identifying individual risk accurately across their population so that they can target effective interventions.

Current methods of assessing risk of care home admission rely largely on face-to-face assessments.  This project will explore the feasibility of using pseudonymous routine electronic data instead. Specifically we shall investigate whether it is possible to predict admission to a care home and high social care costs.  Analogous models have been in use by the NHS in England since 2006 to predict which patients are at risk of emergency hospital admission or readmission.  Patients identified using these models (known as PARR and the Combined Model) can then be offered targeted, preventive care to mitigate this risk.  If equivalent tools could be developed for social care then they might be for both targeting preventive interventions and for setting personalised budgets for direct payments.

As was the case for PARR, the modelling for the current project only represents only a modest proportion of the total work involved—particularly since this project involves the added complexity of working with both health and social services data.  The main phases of the project are:

• Seeking ethics and research governance approvals
• Selecting and liaising with PCTs and CSSRs
• Linkage, pseudonymisation and encryption of data
• Data extraction
• Analysis and modelling

We know from the literature that the explanatory variables for care home admission include, for example, age, sex, ethnicity, deprivation, morbidity, health service use, drugs prescribed, as well as patterns of social care needs and usage.  Clearly these span both health and social care records, and so a complicating yet critical pre-requisite for this project will be to link health and social care data at an individual level whilst securing confidentiality. We are now have strong evidence that routine electronic health and social care data do indeed contain many of the variables that the literature suggests are predictive of care home admission, and we have been able to link health and social care records successfully in a way that protects individuals’ identities.

We have now developed a number of prototype predictive models for social care, which are currently being reviewed by the Department of Health.   We hope to publish our findings later in the Spring of 2010.  Overall, our models have relatively good predictive power (i.e. acceptable positive predictive values) but their ability to identify high risk people across the population is relatively poor (i.e. low sensitivity)   We think this is because the number of social care events in which we are interested are be relatively low.   For example, the PARR model was built using five years’ worth of data on 5 million people, so it contained tens of thousands of hospital admissions.  In contrast, the data available to us for this feasibility project were several orders of magnitude smaller than those used in PARR, so the predictions we have been able to make at this feasibility stage were less sensitive.  Nevertheless, we think they may have a useful role to play in the targeting of early interventions for health and social care.

Please click here for details of Developing a model to predict the use of social care- a paper written by Geraint Lewis, Theo Georghiou and Martin Bardsley, published in the Journal of Care Services Management (23 Dec 2008).

last updated 27/01/2010