Predicting need for intensive social care at individual level

Predictive models are increasingly being used in health care to identify people at high risk of unplanned hospital admission, so that preventive care can be effectively targeted. A study by the Nuffield Trust has now looked at the feasibility of constructing a risk model that could be used in social care to predict an individual person’s future need for intensive social care.
The study obtained routine individual-level data from five sites in England (four primary care trusts and their local authorities, and one care trust). The data spanned several years and described the individual health and social care needs of the people living in these areas, and their use of health and social care services.

The data was examined to see whether prior health and social care information could be used to predict the start of ‘intensive social care’ funded by the council. Intensive social care was defined as a move into a care home, the start of ten or more hours of home care per week, or an increase in annualised social care costs of over £5,000 per year.

Whilst it was possible to construct stable models to predict the start of intensive social care, the models were relatively insensitive: that is, they only detected a small proportion of the people across the population who did start intensive social care. More accurate predictions were achieved when the definition of intensive social care was broadened to include annualised social care costs of above £3000 or above £1000. Significant predictor variables included age (in particular 85+), gender (female), prior social care us, emergency encounters with health services.

Interestingly, models built from social care data alone performed roughly as well as those that contained health and social care data. Nevertheless, certain health variables were significantly predictive of future social care costs.

The study concludes that whilst the predictive accuracy of the models is comparable to some of the models used by the NHS to predict hospital admissions, it is less clear how they might be used in practice. There is a need to pilot and evaluate the tools in a range of sites to see how the models might fit into everyday working practice.

The study also highlighted the long standing issue of promoting and facilitating better data linkages between health and social care data, to benefit not just predictive modelling but also more broadly commissioning of integrated health and social care.

Click here to view a copy of the study by the Nuffield trust 

Case studies
Data analysis
left-behind areas
OCSI news
Project Spotlights
Resources and data