Predicting the future of predictive risk

Blog post

Published: 14/07/2011

Our recent conference on risk adjustment focussed on applications where the stakes are highest – namely risk adjustment linked with the funding and reimbursement of health care services.

The morning session included a description of the work on person based resource allocation (PBRA), with Professor Peter Smith (Imperial College London) and Robert Shaw (Department of Health) presenting. Current work on a PBRA formula for the NHS in England will see risk adjustment being used to set GP practice and possibly clinical commissioning group budgets, based on the predicted per capita inpatient and outpatient costs of their registered populations. Read more about this work on our dedicated project page.

Whilst the PBRA approach was originally applied to give guidance to primary care trusts (PCTs) when setting practice-based commissioning budgets, the same approach could also be applied for clinical commissioning groups (or whatever they are going to be called next week...) as they are based on lists or registers of people rather than being explicitly linked to a geographic area. Don’t miss our new animation for a two minute explanation of PBRA.

The afternoon sessions drew on experiences of applying risk adjustment in other countries, describing the benefits and pitfalls. In terms of controversies, looming over all were the twin perils of:

  • Cream skimming: where an organisation selectively recruits people whose likely costs will probably be less than the allocation they receive, and: 
  • Dumping: where organisations find ways to offload patients whose costs exceed the income.

But we also heard about some of the issues around variable quality information and changes in the completeness of diagnostic coding.

For me, the key issues are not about whether one risk classification system is better than another but more about the ways that these things are implemented.  The empirical performance of a risk adjustment model – which is usually measured imperfectly with a statistic called the r-squared – is only one part of the equation.  

So, for example, in PBRA, although the models' performance would be better if measures of prior health care use were included, these variables were purposefully left out in order to avoid creating perverse incentives. The requirement for simplicity and transparency in the models was also described – Professor Ian Duncan (University of California, Santa Barbara) noted that a number of proprietary models can be something of a ‘black box’. Watch a video interview with Professor Duncan

Perhaps most important though are the issues around risk sharing and questions such as:

  • When costs exceed allocations, how do you determine if this is because of a less efficient care package or simply due to random effects?
  • What exclusions are there to make sure that a commissioner's budget is not at risk from a small number of very high cost patients? How do practices or clinical commissioning groups share these risks?
  • Do you need to allow for the fact that some organisations might want to take on more risk?
  • How do you balance prospective ('ex ante') allocations with retrospective adjustments ('ex post') at the end of the year and how do you make the latter explicit?
  • How do you decide what is a reasonable level of surplus ('profit') to deficit ('loss') to make on an allocation?

There is a sense that we are at the start of a new era of resource allocation models in England. Though the current PBRA models are based on hospital information (currently the most comprehensive source of available data), it is clear that new information streams will become possible in the medium term. 

For example, access to GP clinically coded data is already possible, and schemes such as the GP Extraction Service (GPES) could yield better national models.  

Finally, work on population-level health status measures (through QOF or PROMS), offers another opportunity to test what these models should really include.

All the highlights from the Nuffield Trust’s conference on risk adjustment are available from the dedicated event page. Among the highlights are video interviews with a selection of the keynote speakers, including: Professor Wynand van de Ven, Erasmus University Rotterdam; Professor Ian Duncan, University of California, and Founder of Solucia Consulting; and Dr Dirk Göpffarth, of the German Federal Social Insurance Office.

Suggested citation

Thorlby R (2011) ‘Predicting the future of predictive risk’. Nuffield Trust comment, 14 July 2011.