Tools to help CCGs take predictive risk modelling forward
Predictive risk tools are more efficient and accurate than clinicians at targeting at-risk patient groups – so CCGs should not shy away from them, says Ian Blunt
It is estimated that more than 80 per cent of primary care trusts have used a predictive risk tool since the Patients at Risk of Re-hospitalisation software became available in 2007. Last summer the Department of Health recommended commissioners to look to the market for their next risk tool. As PCTs prepare to hand over to clinical commissioning groups, we look at the market and perceived barriers to effective implementation of such tools.
Predictive risk modelling is a way of “case finding” patients at high risk of a certain outcome – commonly, unplanned hospital admission – who are then offered interventions that may avoid the outcome. Put simply, if you have a list of patients and a list of interventions, a predictive risk tool lets you match them up in the most effective way.
Predictive models should be seen as one part of a wider strategy to manage patients with high needs and it is important to remember that predictive risk tools are only as effective as the intervention they are used to trigger. Increasingly, interventions are proven to be effective but they must be targeted at the right patients.
A predictive risk tool has three parts: the predictive model; the software on which it runs; and the data the model analyses. CCGs can buy whole packages or assemble separate parts. Many models are free but the cost of software or the feeding in of data must be factored in.
While some CCGs might find it easier to outsource implementation, they must be clear about what the model should predict and how it fits in with their overarching strategies.
There is now a wide range of models for predicting various outcomes, including unplanned admissions, readmissions over different timeframes and admissions to residential care. Others are for specific patients, such as people with diabetes. The range, from academic groups and proprietary information tools, is growing all the time.
Companies have developed their own tools which, along with technical support in marshalling data, they supply to organisations. Some business intelligence packages now come with predictive risk modelling built in.
A number of PCTs also developed their own predictive models or software platforms on which to run existing models. Building a predictive risk tool in-house might be a possibility for larger CCGs or their analytical support organisations.
The potential savings from using predictive risk tools are attractive. Emergency hospital admissions have been estimated to cost the NHS £11bn a year – at least £1.4bn of that is for conditions known to be preventable by primary care. Yet there appear to be two main barriers to CCGs embracing predictive risk tools.
There is a risk that GPs leading the CCGs aren’t interested; they would prefer to rely on their own expertise as to which patients need interventions and may have had bad experiences with predictive risk tools in the past. Pointing out that predictions based on clinical opinion may be no more accurate than chance is as likely to offend professional pride as much as encourage use of more reliable methods.
There are opportunities here, however – predictive risk tools can scan whole populations and use data from all patient/health system contacts in a way no single person would be able to do. Flaws of previous attempts (not predicting what clinicians need to predict, too difficult to use or using old data) can be avoided if clinicians are fully engaged with the tool.
The second barrier is that predictive risk tools can require some intensive data wrangling and the provision of analytical support to CCGs is currently uncertain. This might be less of a challenge than it used to be; many PCTs have created data warehouses so the CCG is more likely to find the necessary data is linked and ready to use.
Many believe the future of healthcare is integrated care underpinned by integrated IT systems – perfect territory for predictive risk tools. As more data become available, it will become easier to produce predictions and those predictions will becomes more accurate. If CCGs engage with predictive risk modelling, are clear about what they want their models to predict and how they fit into their overarching strategies
then they can help to drive this development forward.
Ian Blunt will speak at the Commissioning Show on 27-28 June.