Clustering around in Mental Health

A great conference last week on the introduction of new payment arrangements for NHS Mental Healthcare! And it was all the more interesting because, far from being exclusively limited to Mental Health, much of what it had to say was relevant to significant areas of Acute care too.

For anyone who has not been following these developments closely, from 1 April it becomes mandatory in Mental Health to start using so-called ‘Clusters’ to classify cases, and include them within the process of contracting for care. There is general recognition that we are still a long way from being able to base budgets exclusively on Clusters, let alone set a national tariff for them: there is simply too much variation among cases within Clusters, or more precisely too much unexplained variation between them, to allow them to be used reliably in this way.

On the other hand, it is a breakthrough that large numbers of Mental Health Trusts are at least applying the Clusters to their records. One of the conference speakers mentioned that his Trust was achieving nearly 98% coverage, though he was candid enough to admit that he had his own doubts about the figure: in relation to just what should he be measuring the percentage? There is a grey area of definition between what should be included in Clustering and what should be excluded, so getting a precise percentage is difficult.

Even so, it’s clear that a high proportion of cases are now being Clustered, and that’s a major advance. It means that we can at last begin to see just how well Clustering is working, to assess the level of variability and, ideally, to work out what is acceptable variation and what is not. In particular, we need to find out where variation is simply down to the way Clustering is being applied. 

Because Clustering isn’t like HRG or American DRG grouping, and not just because there are only 20 or so Clusters as opposed to nearly 1200 HRGs. The biggest difference is that HRGs can be derived by an automated system based on data already recorded against patient records — diagnoses, procedures, lengths of stay, etc. — whereas Clusters are based on a professional’s assessment of the case.

This makes the extent of variation in treatment within a single Cluster unsurprisingly high. For example, another speaker reported on the results of a survey of over fifty Mental Health Trusts. Within Cluster 11 alone, the cost per day of treatment varies from Trust to Trust from a few pounds up to nearly £550:

Cost per day variation across Trusts — within a single Cluster

From the graph, it look as though between about Trust 11 and Trust 35, daily costs seem to be in the £20-£30 range, suggesting a reasonable level of consistency. Outside that range, however, variability is so high as to undermine the system: at the ends of the distribution, it is of the order of 100:1 or more.

There are at least three possible explanations of this variation:

1. Wrong Clusters: the clinician’s assessment is incorrect. In this context, a speaker at the conference mentioned the ‘Richmond-Lambeth’ syndrome: mental health problems tend to be far more pronounced in the under-privileged London borough of Lambeth than in relatively well-heeled Richmond; will that lead psychiatrists in Richmond to include in the more severe Clusters service users who may be more seriously ill than many others of their case mix, but far less than those included in similar Clusters in Lambeth? Even without such general trends, it is of course possible that individual cases can slip into an inappropriate Cluster.


2. Poorly-defined Clusters: some of the Clusters may be too broadly defined and therefore cover cases that are not entirely homogeneous, but include service users whose condition differs too much in severity for their treatment to be comparable.


3. There are genuine variations in clinical practice within a Cluster of reasonably homogeneous service users.

In the longer run, it is only the third type that is of interest: we want to identify, analyse and take action over variations in practice that can’t be justified by any specific characteristic of the service user.

On the other hand, in the short term it is certainly the first two that are going to attract most of our attention. Until that kind of problem can be ruled out as a possible cause of the differences we see between cases and between providers, we can’t really use the data for analysing the third type of variation. And, above all, we certainly can’t use the Clusters as a reliable guide to cost.

So the first stage of the exercise is going to be looking into what exactly lies behind each of the Clusters and what is causing the observed differences. As we do that, we shall start to build a picture of what we would normally expect to see in the way of a mix of treatment types within a Cluster: between so many and so many outpatient appointments or community visits, between so many and so many admissions or days of inpatient care.

In other words, we shall start to build definitions of the packages of care that are associated with Clusters. When we have those, we shall be able to identify treatment profiles that differ significantly from the norm.

Now there’s nothing exclusive to Mental Health in this approach to defining bundles or packages of care. We can build them for Mental Health care clusters, but why not for somatic diseases too? Why not for congestive heart failure, diabetes or even different types of cancer? Indeed, for any long term condition?

Because this kind of work is breaking down another of the deeply-established barriers in healthcare, between somatic and psychiatric care. Anything that requires treatment over a long period, in different care settings, perhaps by different providers, lends itself to this kind of package of care approach. It’s by no means limited to Mental Health only.

Nor is it limited to British healthcare only: much of the thinking behind the launching of Accountable Care Organisations in the United States is also concerned with having a single body responsible for care in a variety of settings by a range of different institutions. There is nothing surprising about this convergence: it is a piece of increasingly accepted wisdom that anything up to 40% of what we previously thought of as acute care is evolving into chronic condition management (with Cancer as perhaps the most striking example). Inevitably, that is driving us all, everywhere, to undertake this kind of work.

So again it was interesting to discover that many of the people attending the Mental Health conference also had responsibility for helping to manage Long Term Conditions. The need to think in terms of packages of care spans traditionally distinct fields.

Interesting times ahead. And, not for the first time, I was struck by how Mental Health is showing the way.

There are some interesting challenges ahead. Not least is what lay behind what several speakers pointed out: they didn’t like talking about Payment by Results. They felt that the initials ‘PbR’ should be viewed as standing for ‘Payment by Recovery.’

A refreshing view, and one that fits well with the package of care approach. After all, how do you know a package is complete except when the patient has recovered? 


But can you imagine the impact on healthcare if remuneration started to be based on outcome?