Predictive modelling – enthusiasm versus pragmatism

Blog post

Published: 27/06/2012

It was a privilege to present the progress we’ve made in Devon at the predictive risk 2012 conference. Both Todd Chenore and I appreciated the chance to outline our results of integrating predictive modelling into a whole system approach, which appear to be consistent and transferable to other areas.

Predictive modelling and virtual wards in Devon began with three practices in North Devon in 2008 and the service now covers three distinct localities across Devon.  They have implemented the service in three slightly different ways that appear to have had a direct impact on the outcomes seen. 

The best performing (population 220,434) is showing a 22 per cent reduction in admissions from the top 0.5 per cent of the population (comparing 2009 to 2011) through higher targeting of the virtual wards on this high-risk group and providing ‘on the ground’ project support.  This is an estimated saving of more than £845,000 on previous years’ figures (excluding any growth).

Key to this is a consistent and focussed application of the intervention. Ensuring that once the high risk patients are identified, the process cycle keeps turning – with monthly risk-profiling and patient turn-over – and keeping a lid on the potential for this changing group’s admissions to flare.

Overall, the drive is towards gaining consistent control of the current and impending demands. The next challenge is moving the identification further upstream and improving the responsiveness of the system. This obviously has its challenges when the intervention is a multi-disciplinary (and therefore multi-provider) service with priorities that can sometimes be at odds.

In Devon the sign-up to the predictive modelling accelerated significantly with a Commissioning for Quality and Innovation Locally Enhanced Service (CQUIN LES) payment to practices.  Many areas appear to be using financial incentives, and the various discussions during the conference confirmed this as a common and key driver for change that has helped achieve a critical mass in many areas.

When rolling out at scale the issues are greater, with exponential increases in hearts and minds to win over, agreements required, integration to achieve, data governance etc.

It is certainly not a simple thing to do. It was interesting to see how the US and Germany, with their contrasting systems, are also seeing the benefits of the approach.

The impact of individuals in the process is also key. Enthusiasts, keen to make change and embrace new concepts can drive these projects.  Conversely, you can have a few influential individuals who can block projects from gaining momentum.

Cautious pragmatism should always be there but in some situations it appears the perception that “clinician knows best” still purveys. And despite the emerging evidence, the concept of predictive modelling to some remains a black-art or an unsubstantiated fad.

The potential though is difficult to ignore, especially with the apparent limited options available to prevent a bleak future ahead for the NHS.

The availability of predictive modelling is also at a turning point.  I personally couldn’t understand the decision not to centrally provide an update to the Combined Predictive Model (CPM)

We’ve demonstrated in Devon the ability to implement and locally calibrate/improve a predictive model derived from the CPM on a medium to large scale (1.1 milllion patients). It is therefore frustrating to see many purchasing third-party models at six-figure costs.

I would encourage any clinical commissioning group (CCG) thinking of doing this to look at the options locally. I hope the Department of Health reviews this decision whilst commissioning support is still under the NHS umbrella.

Overall, the results we’re seeing in Devon, the continuing enthusiasm of many and the obvious need for the NHS to modernise its approach all seem to point to a promising future. A future where predictive modelling and targeted, multi-disciplinary, integrated case-management will eventually become the norm in the UK.

Dr Lovell is GP and Urgent Care Lead at NHS Devon. Please note that the views expressed in guest blogs on the Nuffield Trust website are the authors' own.

Suggested citation

Lovell P (2012) ‘Predictive modelling – enthusiasm versus pragmatism’. Nuffield Trust comment, 27 June 2012.