Predictive risk: an idea whose time has come?

Blog post

Published: 29/07/2013

The Nuffield Trust recently held its fourth annual conference on predictive risk - or applying statistical models to populations in order to identify patients who might benefit from health interventions of various kinds.

The mere fact that this was the fourth conference on the subject shows that this is an idea with legs. And, indeed, it has just been given a boost by the Department of Health, which has included a directed enhanced service for ‘risk profiling and care management’ in the latest GP contract.

Dr Geraint Lewis, an early pioneer in the area of risk profiling and developing tools for the job, is now Chief Data Officer of NHS England.

Today’s predictive risk models are built around data from the Hospital Episode Statistics, and Dr Lewis was able to tell the conference that NHS England is planning to expand HES, while linking it to data from GP, mental health and community data sets in the new service.

At the moment, the evidence suggests that many programmes improve patient satisfaction and the quality of the service they receive; but do not save money

This should open up exciting new opportunities for researchers; although another speaker, Professor John Billings of New York University, suggested that it might not make too much difference to the most widely used models, which try to spot people at risk of unplanned admission to hospital.

Dr Lewis also indicated that patients would also be able to download their histograph (although it wasn’t quite clear why they would want a one-off snapshot of such complex data, when they are being promised access to their GP records and more useful services, like online booking).

On the downside, clinical commissioning groups and their commissioning support units are finding it difficult to operate some predictive risk tools at the moment.

Unlike the primary care trusts that preceded them, CCGs do not have the legal right to handle patient identifiable data; and they are not likely to get it. Dame Fiona Caldicott’s second review of information governance concluded that pseudonymised data should be sufficient for commissioning.

Conference speakers could only advise commissioners, their support services, and companies wanting to work with them to consult NHS England guidance on staying within the law. Dr Lewis also promised that some kind of patient information poster campaign is being devised (although goodness knows how predictive risk will be summed up for a poster).

Once patients are identified, suitable interventions have to be devised for them, and much of this year’s conference was taken up with discussing how to evaluate whether these work.

Sophisticated methods are needed to identify a control group for comparison, and timing is key; some of the early evaluations have stopped after a year or so, when projects are just getting going. This may be one reason why, at the moment, the evidence suggests that many programmes improve patient satisfaction and the quality of the service they receive; but do not save money.

Interestingly, Dr Sarah Purdy from the University of Bristol said a meta analysis also suggested that good access to ‘old fashioned’ GP care might have more of an impact than ‘flashier’ interventions, such as telehealth monitoring.

This may be a hard sell for politicians and policy makers, who like to be seen promoting innovation. So it needs further exploration; and will no doubt be a subject for further conferences.

Lyn Whitfield is a Journalist and Editor at E-Health Insider. Please note that the views expressed in guest blogs on the Nuffield Trust website are the authors’ own.

Suggested citation

Whitfield L (2013) ‘Predictive risk: an idea whose time has come?’. Nuffield Trust comment, 29 July 2013.