Risk model applied to social care
Researchers at the Nuffield Trust have succeeded in building a model to identify people at risk of needing social care – but admit there would be issues with using it in practice.
The researchers, many of whom worked on predictive risk models for the NHS when based at the King’s Fund, recommend careful testing, piloting and evaluation of the model’s impact on health and social care costs before it is widely used “in the real world.”
However, Dr Martin Bardsley, head of research at the Nuffield Trust, said: “Linking patients’ information in this way has real potential to improve the quality of care patients receive.
“The prize is not only greater independence for older people, but also significant potential cost savings for health and social care budgets – critical as UK public services enter a period of constrained funding.”
The Department of Health commissioned work to develop three models to identify patients at risk of emergency readmission or admission to hospital, who might benefit from intensive support.
In 2006, the Department for Communities and Local Government commissioned a study to explore whether a similar model could be created to identify people at risk of admission to a care home.
The researchers concluded that this should be possible in theory, after which the DH commissioned the Nuffield Trust to try and build one in practice.
The researchers used pseudonymised data from social, primary and secondary care from five communities in England to try and identify patients at risk of admission to a care home within 12 months.
Only a very small number of people in a given population will need to go into care within a year. As a result, the researchers found it hard to pick them out. The most sensitive model they built picked up only one in five (its sensitivity was 17%).
On the other hand, the model was accurate in the sense that the people it identified as being at risk of going into care were likely to end up in care (the positive predictive value was 50%).
“Had we simply tossed a coin for each person aged 75 and above [to decide whether they would go into care or not, it would have performed with] a sensitivity of 50% and PPV of about 1.2%. So our models are about 30 times better than chance,” the researchers write in their report.
The PARR model used by the NHS to identify patients at risk of emergency readmission to hospital is much more sensitive – it picks up about two thirds of the people it is looking for – and has about the same level of accuracy.
Dropping the threshold from being at risk of going into a care home with 12 months to needing social care costing £5,000, £3,000, and £1,000 picked up more people. But providing support for them might be less cost effective.
Somewhat surprisingly, the researchers say some of the councils they worked with are still interested in using the lower-threshold models to “implement very low-cost interventions (posting brochures and information leaflets, for example).”
Others wanted to immediately use the model in its original form to offer high levels of support to individuals who might cost them £30,000-£40,000 a year if admitted to a care home.
The researchers point out that this would present data protection problems as things stand.
It would also raise “ethical concerns in terms of the justice of spending large amounts of resources only on people identified by the model” when so many others would go without.
To improve the sensitivity of the model, the researchers suggest trying to include more data, including information from people spending their own resources on care and support, who may appear “out of the blue” when their money is exhausted and they qualify for council care.
Another key recommendation is that social care data should be more consistently recorded, with a view to creating a social care database analogous to the Secondary Uses Service and Hospital Episode Statistics in healthcare.
Report: Predicting Social Care Costs: a feasibility study. The Nuffield Trust.
Last updated: 23 February 2011 17:27
© 2015 Digital Health Intelligence Limited.