In their piece on mental health Payment by Results, ^{Reference Macdonald and Elphick1} Macdonald & Elphick note ‘a lack of reassurance that costs per case within a cluster will be similar enough to support a viable tariff calculation’. This may underestimate the difficulties with the proposed new payment mechanism, which may have effects wider than disruption of nascent routine outcome measurement systems.

The UK has come relatively late to the process of payment reform for mental health services, but despite this it has followed an approach unlike that in other countries. The fundamental principle behind the care cluster approach seems to be the presumption that individuals with similar needs for care, as notionally defined by clusters of scores on the Health of the Nation Outcome Scales (HoNOS), will receive similar care and therefore incur similar costs. Importantly, costs themselves did not enter into the original process of defining care clusters. ^{Reference Self, Rigby, Leggett and Paxton2}

The approach pioneered by Fetter at Yale ^{Reference Fetter3} in developing the original Medicare prospective payment system in the 1980s was to combine consultation with clinicians and statistical analysis of clinical, administrative and cost data using variance reduction so that case-mix groupings are both expected to produce similar ‘clinical responses’ and also do in fact demonstrate acceptable homogeneity of costs. This approach was also followed by Australia and New Zealand, ^{Reference Buckingham, Burgess, Solomon, Pirkis and Eagar4,Reference Eagar, Gaines, Burgess, Green, Bower and Buckingham5} when they attempted to develop payment systems based on HoNOS. Achieving homogeneous costs within groups is crucial because it minimises the random risk to providers (the risk that appropriately incurred costs and therefore revenue differ randomly from those reimbursed). A typical cut-off for acceptable cost homogeneity is for each case-mix group to have a coefficient of variation of one or less (mean divided by standard deviation). It is also essential to make sure that factors relevant in resource use which may be more or less prevalent among different providers are also represented; otherwise there may be an element of systematic risk, with certain providers being systematically underpaid and others systematically overpaid. Even when this more standard approach is followed, it may not be successful, especially in mental health, where cost variation is high. Infamously, the original Medicare prospective payment system was never implemented in specialist mental health units in the face of evidence that resource homogeneity was too low and that the system would systematically disadvantage those units, and has now been abandoned in favour of an across the board per diem payment system for psychiatric in-patients. ^{Reference Lave6} Neither the Australian nor New Zealand systems were ever implemented.

In the light of the foregoing comments, it is perhaps not surprising that the Department of Health's own pilot studies from 2006 demonstrate both that resource homogeneity of care clusters is unacceptably low, with only 1 of 13 clusters having a coefficient of variation of less than one, and also that far better homogeneity could have been achieved, especially for in-patients, had the standard variance-reduction approach been followed. ⁷ At present, it seems to me that the lowest risk approach to reforming payment for mental health services is to adopt a basic system of activity-based funding, and use the data collected in this way, along with clinical and administrative data, as part of a slow and careful process of reform. Certainly, payment for mental health services in the UK is ripe for reform, as variations in resource use between providers are far wider than could be accounted for by any difference in case-mix. ⁸ But this may not be the best way of approaching it.