Since the development of the original PARR and Combined Predictive Model tools, many PCTs have introduced these or similar case finding tools, to identify people at risk of unplanned hospital admissions.
We recently held a conference for people interested in using predictive risk tools in health care. People from across the UK and further afield spoke about the ways in which these tools are being used, as well as highlighting some issues and cautionary tales that have come from experience in trying to introduce and use these systems.
I was struck by the sheer range of uses to which predictive risk tools are now being applied. Until relatively recently, tools such as PARR and the Combined Predictive Model were pretty much the only show in town.
Some speakers identified issues with getting clinicians to accept that predictive tools can be a useful aid to decision making and case finding
However, throughout the day we saw examples of people developing tailored models, such as the SPARRA (Scottish patients at risk of readmission and admission) tool, created by the Information Services Division in Scotland, which aims to predict unplanned admissions for different groups, such as the frail elderly, and those with long term conditions. It is even being extended to try to predict emergency department attendances by young people.
Predictive tools that are designed for different purposes was also a theme of the day. We talked about identifying people at risk of developing particular conditions; those who are near the end of life; allocating budgets to healthcare providers for treating people with mental health conditions; and potentially to look for evidence of poor quality care and even fraud detection.
And the list goes on.
This is evidence of a field that is rapidly developing in its scope and ambition. It was good to see people taking existing models (such as the Combined Predictive Model), and tailoring them to their local circumstances. This might involve calibrating them to the profile of use in their local population, but we also saw people building in additional local datasets such as out of hours care.
Developments like these have the potential to improve the validity of models for local populations, and to increase buy-in from clinicians. The question of clinical engagement was one that came up repeatedly.
Some speakers identified issues with getting clinicians to accept that predictive tools can be a useful aid to decision making and case finding. Others talked about their experiences of using financial incentives to encourage GPs and others to take part in the rollout of predictive tools and associated interventions.
However this feels like an area that needs a lot of further work. How do you get clinician buy-in? Are financial incentives the answer (probably not in the long-term)? Is it about evidence, finding local clinical champions, or something else entirely?
Another recurring question was how best to properly evaluate the success (or otherwise) of the interventions that are put in place. John Billings of New York University (and Senior Associate of the Nuffield Trust) gave a fascinating talk about the successes and failures of a project in New York.
They had used predictive risk tools to target people with chronic conditions for case management. He emphasised that making full use of the power of predictive risk tools requires a clear idea of the outcome you are trying to influence (and being confident that you can actually influence it). Then designing a proper intervention and fully evaluating it
It was clear that there is growing interest in predictive risk tools. However, the questions of clinical scepticism, the incentives to encourage participation, as well as the relative paucity of well-designed evaluations all mean that significant hurdles still lie ahead before these tools become standard.
Xavier Chitnis is a senior reseach analyst at The Nuffield Trust.
This article first appeared on the blog section of the Nuffield Trust website. www.nuffieldtrust.org.uk