The government's primary care reforms mean that resources must be shared out according to a new unit of analysis - the primary care group. How will this be done? The current proposal is to apply the national resource allocation formula, which produces weighted capitation target payments derived from socio-economic and morbidity data from the 1991 census.
These payments are intended to represent health authorities' and PCGs' fair share of available resources based on the health needs of their populations.
The formula aims to ensure that funding is allocated on the basis of the size and age of the population, its health and socio-economic needs, and the unavoidable costs of providing healthcare across the country.
Data difficulties There are some problems with the formula, however. PCGs are defined by groups of GP patient lists. Census variables and mortality data are not collected for GP practice lists (or groups of lists) but instead are based on enumeration districts or electoral wards.
Some mapping, attributing the characteristics of the larger populations - the districts and wards - to the smaller populations of PCGs and general practices, is necessary at HA level. But this is more of an issue at GP practice level since GP lists are made up of patients from many different districts and wards. The accuracy of attribution is crucial to the working of the resource allocation formula at this level.
Attribution is about giving each individual registered to a specific GP practice the characteristics (as measured in the 1991 census) of the area in which they live. But this means it may be subject to the 'ecological fallacy' - errors may occur when generalising from a population to an individual, since an individual may not be representative of the area in which they live. This could mean that the 'fair share' estimations derived by the national formula may not have the intended impact on equity.
Given these concerns, evidence is required on the accuracy of attributing data from electoral wards and enumeration districts to GP practices and PCGs to assess whether the national weighted capitation formula can be used to allocate resources to these smaller units.
Attribution in practice
Research was carried out in 1998 looking at 199 GP practices in Northumberland, Camden and Islington, and Doncaster HAs.
1 This compared actual demographic characteristics of practice populations with those which might be expected on the basis of attribution of 1991 census demography.
For a range of demographic variables, such as the proportion of the population in certain age bands, the mean attribution error was calculated (see box, below). Several factors affecting attribution accuracy were then analysed.
Accuracy and error
Attribution error was found to be slightly lower overall for enumeration districts, compared with electoral wards. For example, the value for the percentage of the population aged over 74 estimated by attributing from the enumeration district differed overall by an average of 26.4 per cent from the actual values observed on each practice list. This compared with 27.3 per cent when attributing from the electoral ward data.
The characteristics of demographic variables were also examined. It was found that for 'smaller' variables, covering a smaller proportion of the total population, such as those aged over 84, attribution error was higher than for 'larger' variables.
This is intuitive, as with smaller variables a single individual represents a larger proportion of the whole, so each one constitutes a high error.
Variables with greater variability were associated with lower attribution accuracy.
The impact of changes in the population on attribution accuracy was also considered. Since the 1991 census Camden and Islington HA has seen a large increase in population, and thus a big increase in the size of local GPs' lists - around 28.3 per cent.
Overall attribution accuracy of demographic variables is poor. For example, for the percentage of the population aged 0-14 years, attribution error was 41 per cent, while for the percentage aged over 74 years it was 81 per cent.
Doncaster HA has had positive natural growth and negative net migration, and GP list growth is 3.8 per cent. Attribution error is much lower than in Camden and Islington. Northumberland HA has had neutral population change and list growth is 2.8 per cent. It also had the smallest attribution errors.
Generally, attribution accuracy was found to be best with stable populations. Deterioration in attribution performance is particularly pronounced for variables that include young and elderly people, where natural population change has been greatest.
The risk of ecological fallacy declines the higher the percentage of individuals drawn from a particular enumeration district or electoral ward, since there is reduced risk of these individuals being unrepresentative. A new measure of 'zonal concentration' of GP and PCG lists was devised to capture this effect. The correlation between zonal concentration - the proportion of a ward population accounted for by a GP practice list - and attribution error was also examined. If a GP list makes up 90 per cent of a ward, it is very concentrated in one zone. The practice population is very similar to the ward population so attribution is more accurate. Results showed that attribution error is indeed lower where there is a higher zonal concentration for almost all variables.
Because the national formula includes composite variables, made up of individual census variables multiplied together, the impact of composite variables on attribution accuracy was also examined.
For the three HAs the attribution error of a composite of three uncorrelated demographic variables was calculated.
Individually the errors for the three variables are 18.5 per cent, 50.9 per cent and 34.8 per cent, giving an average of 34.7 per cent. But when calculated as a composite variable, the attribution error was much lower at 10.4 per cent.
So what does this mean for allocating resources to PCGs? Hypothetical PCGs with patient lists of 20,000; 60,000; 100,000 and 130,000 were constructed for Doncaster HA, as well as groups based on average GP practice lists (5,500). Data was aggregated from HA information into eight age-band variables. Attribution error was calculated for each variable.
Figure 1 shows that attribution error (Northumberland HA) falls markedly from list size 5,500 patients to 20,000 patients, with further improvement at 60,000 patients, and marginal thereafter. While the trend is similar for all variables, figure 1 also shows some differences . There is a mean absolute attribution error of 13 per cent for the composite variable at practice level and just 3 per cent for PCGs with a population of 100,000.
As PCG size increases, zonal concentration also rises, as figure 2 shows. This is the key reason for the enhanced performance of attribution.
So this work shows that errors are large at practice level but decline markedly for population sizes that reflect the size of PCGs.
Increasing list size improves the accuracy of attribution and this is mainly due to increased zonal concentration. This means that attribution is likely to be suitable for mapping to PCG population, but not individual practices.
Resource allocation for PCGs is an important issue since they have no historical trend by which to set budgets and there is a chance to establish 'fair share' allocations from the onset.
The errors associated with attribution at PCG level are small - but remain undesirable. At the moment, however, there is no alternative to its use in ensuring an equitable allocation of resources. No data is available at practice level that can be used to assess health need as a basis for resource allocation. The costeffectiveness and implications for equity of collecting practice level data needs to be considered.
Equation for calculating mean attribution error (actual practice value - attributed practice value) / actual practice value x100
The proposed formula for allocating funds to primary care groups using census data raises concerns about equity.
The proposed formula is based on health needs, but these are difficult to define at practice level.
Research applying the proposed formula to 199 practices showed it was more suitable for PCGs than individual practices.
Richard Cannock is a former NHS Executive assistant economist working on resource allocation issues for GPs and PCGs. Paul Miller is lecturer in health economics , Trent Institute for Health Services Research, Nottingham University.
1 Cannock R. Can We Use the National Weighted Capitation Formula to Allocate Resources to GP Practices and Primary Care Groups? Paper presented at the Health Economist Study Group meeting, Galway, Summer 1998.