Resource allocation to health authorities: the quest for an equitable formula in Britain and SwedenBMJ 1997; 315 doi: https://doi.org/10.1136/bmj.315.7112.875 (Published 04 October 1997) Cite this as: BMJ 1997;315:875
- a Department of Public Health Sciences, Division of Social Medicine, Karolinska Institute, S-172 83 Sundbyberg, Sweden
- Correspondence to: Professor Diderichsen
- Accepted 6 March 1997
In recent years countries with very different healthcare systems have been showing increasing interest in resource allocation policies based on weighted capitation. In countries whose healthcare systems have competing health insurers the main concern has been to construct capitation formulas that prevent favourable risk selection or “cherrypicking.” Reforms to the American Medicare programme and Dutch healthcare proposals have stimulated renewed efforts to find a way of overcoming this problem.1 2 3 4
Countries with national health services, such as the United Kingdom and Sweden, have also experienced far-reaching reforms of health care, with important implications for equity in access to care.5 6 Risk selection should be less of a problem, at least with health authority purchasing, as the population is assigned to a purchaser based on area of residence. The new role of local purchaser, however, calls for more exact methods to allocate “purchasing power,” because local areas will show stronger variation in relative need than regions and counties.
We outline British experiences in attempting to devise an equitable formula then present the new model that we have developed in Sweden for Stockholm County Council. We discuss what lessons these experiences hold for other countries facing a similar challenge.
In Britain serious attempts to devise more equitable mechanisms for resource allocation for the NHS date back to the 1970s, when it became clear that funding to the regions based on historical activity had perpetuated the inequalities in funding that existed before the NHS. Since then, development work has gone through three distinct phases.7
In the first phase the formula created by the Resource Allocation Working Party was developed for distributing resources from central government to regions. It used mortality in each area as an indicator of healthcare need.8 The formula was in use from 1977–90 and gradually managed to redistribute resources from the metropolitan regions to the poorer regions in the north.9
In the second phase the argument that the measurement of need should be based on empirical data led to a new formula for weighted capitation, applied from 1991 to 1995.10 This empirical approach was severely criticised on methodological grounds and because it seemed inequitable.11 12 13 14 15 16
The United Kingdom and Sweden face similar problems in how to achieve a fair allocation of resources within a purchaser-provider system
In contrast with the British formula, the new Swedish approach is based on individual level data and uses demographic and socioeconomic variables as proxy measures of healthcare need
The Swedish model incorporates actual, rather than estimated, costs of care
The resulting model allocates proportionately more resources to populations with poorer health and socioeconomic characteristics
Both the Swedish and British approaches illustrate the practical problems and the highly political nature of resource allocation
These experiences hold important lessons—not least for the growing number of other countries with a similar quest
Clearly, the Department of Health needed a more sophisticated model for allocating funds directly to local districts now that they were purchasers. It commissioned health economists at York University to develop a more sensitive, empirically based model, to be incorporated into a third allocation formula from April 1995 onwards.
The York model is based on an ecological study of small areas to identify the determinants of use of hospital services.17 18 The need variables identified include both health and socioeconomic factors (table 1). In addition, statistical models were developed to distinguish several confounding influences on the use of services, such as the supply of hospital beds and general practitioners. The effect of applying the formula in full at the district level would be to redistribute funds towards poorer, inner city areas.18 The Department of Health decided, however, that the full York model would apply to only 76% of funding and the new arrangements would be introduced only gradually over several years. Other adjustments for “market forces” were also added. In effect, these adjustments watered down the full potential of the York model to allocate resources equitably. As about 70% (£23bn a year) of NHS funding is distributed through these formulas, even slight adjustments can make a big difference to local allocations.
New approach in Sweden
Like Britain, Sweden has a national health service, publicly funded and provided. Of the total healthcare budget of Kr82bn (£8bn), 82% comes from regional income taxes raised by the 26 county councils responsible for administering health care.
This regional funding has until recently been distributed directly to public hospitals and primary care centres on the basis of historical activity, adjusted for inflation. This has changed in the past four years in counties that have introduced an internal market. In particular, Stockholm has been at the forefront of the introduction of a purchaser-provider split, and associated developments in resource allocation have consequently gone further than in the other counties.
Stockholm County Council serves a population of 1.7 million with a healthcare budget of £1.6bn. Most (90%) of the county budget is distributed to nine health authorities, each covering populations of between 50 000 and 300 000.
Basis of model
The contrasting features of the Swedish and British approaches are listed in the box. Individual level analysis was chosen not only because of the practical availability of data but also because of the problems inherent in ecological analysis.19 20
Distinctive features of resource allocation in Sweden and Britain
Sweden (stockholm model)
Need for health care is measured by demographic and socioeconomic variables rather than mortality or other health status indicators
Analysis is based on individual level data rather than at a small area (ecological) level
Actual, rather than estimated, relative costs of health care are used
Need is measured by mortality, self reported morbidity, and various socioeconomic variables
Analysis based on an ecological study of small areas to identify the determinants of inpatient services
The estimates are adjusted for the confounding influences of supply on geographic variations in use
Estimated costs of health care are used
Finding a direct indicator of health status for measuring healthcare need that could be linked to individual use of health care and cost data proved difficult. The model therefore uses various socioeconomic indicators as proxies for healthcare need, over and above that created by the demographic profile of the population. The choice was based on evidence showing that use of hospital services in Sweden was proportional to the relative need of major socioeconomic groups.21 22 Higher use by more socially disadvantaged groups is assumed to translate into higher costs of care, for which health authorities need to be funded.
Psychiatric services, however, were used at a low level by non-Nordic immigrants, perhaps not reflecting all their needs.22 Immigrant status was therefore excluded from the analysis. A different model was devised for primary care (not reported here).
The analysis makes use of the personal identification number, which everyone in Sweden has and which can link healthcare records with census and other socioeconomic databases. Since a new system of payment was introduced in 1994, actual costs of care billed to purchasers have also been available for each individual in the population. The analysis has four main stages.
Stage 1—We created two new databases each year, linking the records on healthcare use and related costs to data on age, sex, socioeconomic group, education, cohabitation and marital status, country of birth, and housing conditions. One database covered a 30% random sample of the country's population, containing their socioeconomic characteristics and any health care they had used. The other database included all people with inpatient care and their background variables.
Stage 2—We then tested different models (with multivariate Poisson regression of outpatient and inpatient episodes) to select the demographic and socioeconomic variables that had the greatest effect on use, controlling for other variables. The variables selected by this process for the final model were (a) age in 10 classes; (b) socioeconomic groups in four groups based on occupation and employment (education for pensioners); (c) cohabitation and marital status in four classes; and (d) housing in five classes, according to tenure and size. Sex was not included in the final model. The effect of including sex made a negligible difference to the distribution of resources as the distribution of men and women did not differ between districts.
Stage 3—A matrix was constructed in which each cell represented a unique combination of the selected variables. In each cell, weights were calculated equal to average costs per inhabitant. Separate weights were calculated for acute medical and surgical care, non-acute care, and psychiatric care. Because actual costs were not available for psychiatry, the costs for this specialty were estimated on the basis of number of bed days and outpatient visits. Table 2 shows an abridged version of the matrix.
Stage 4—A corresponding matrix with the number of inhabitants in each of the nine health authority areas was then constructed, and each individual was ascribed a weight based on their social and demographic characteristics. These weighted individuals were then summarised for each area and the budget calculated as a proportion of the total sum for the whole county council (table 3).
The model has been applied gradually in calculating health authority budgets in Stockholm County Council since 1992. Before 1996, costs were estimated from the number of admissions and bed days, whereas the 1996 budget was based on actual costs for the purchasers.
Overall, the model has allocated more resources for the care of people living in more disadvantaged socioeconomic circumstances (table 2). The resulting ranking of authorities in table 3, based on these costs, follows the known differentials in health, demographic, and socioeconomic factors in the county.23
The interim model used in 1995, based on estimated costs, allocated a large share of the budget to areas containing a high proportion of elderly people and people living alone. As the year unfolded, it became apparent that the interim model might have overcompensated for the costs of providing health services for elderly people. In fact, central Stockholm, with the highest proportion of elderly people, could not spend all its allocated budget, whereas the suburban areas with young families ran up budget deficits. When actual costs became available for the 1996 model, it was found that each bed day was cheaper for elderly than for younger age groups. In 1996 therefore the share of the budget was reduced for central and south Stockholm and increased for suburban areas (a shift of 1.4% of the budget) (table 3). Politically, this was seen as too great a shift to be achieved in one year. The county council therefore gave extra funds in the 1996 allocation to the authority hardest hit by the redistribution.
Insights from these developments
What are the lessons from these British and Swedish experiences? In both countries the principle has been firmly established that healthcare resources should be distributed in proportion to the relative needs of local populations. It is a step forward that serious attempts are being made to translate this principle into practice, but the quest for improvements continues.
Making best use of available data
The experiences illustrate two different ways of going about the task, largely determined by the need to make the best use of whatever data are routinely available in each country. This has led to an analysis based on area of residence (ecological analysis) in Britain and an approach based on data from individuals in Sweden. Several commentators have concluded that individual level analysis is the better option, to reduce the problems of confounding and misclassification.24 25 The Swedish approach has made the most of the opportunity offered by newly available individual data, though this was the only practicable option for Sweden because the small numbers obtained from area based data would have made the resulting statistical models unstable. It did, however, restrict the choice of indicators of need. For example, no suitable health indicators were available that could be linked to the other individual level data.
The York model has to rely on data for small areas, not directly linked to individuals, which brings added problems of interpretation. On the other hand, with care it can include additional local data on mortality and morbidity, increasing its sensitivity to geographical variations that are not simply the sum of individual variations in the basic sociodemographic characteristics.
Proxies for need
Both the British and Swedish approaches are based on the assumption that the different needs for health care of the various sections of the population are matched by their differential use of services. But in practice the use of services is influenced not only by legitimate need but also by supply and many other socioeconomic factors, so the match is not perfect. Given the circumstances, informed judgments have to be made on the most practical solutions. The Swedish decision, for example, to leave out an indicator of “ethnic group” from the final analysis was based on the evidence that non-Nordic immigrants have higher psychiatric morbidity but a relatively low rate of use of psychiatric services. Incorporating a factor based on use by ethnic group would have led to fewer resources being allocated to health authorities with large immigrant populations.
Taking deprivation into account
Both approaches consider it essential to take social and material deprivation into account. They have both selected employment factors and living alone as important indicators of increased need for healthcare resources. Sweden has added indicators of poorer housing, and Britain has added households containing singlehanded carers (including single parents) as well as direct health indicators.
Two new relevant findings emerge from the Swedish data on differential costs of care. Firstly, the analysis of actual costs for care of different groups provides a direct demonstration of the higher costs incurred by more disadvantaged groups in the population and the need for extra resources in areas where the proportion of people from these groups is greatest. Secondly, the comparison of estimated costs in 1995 with actual costs in 1996 revealed the scale of the bias introduced when only estimated costs are used. A similar problem with estimated age-cost weights was encountered in the British formula introduced in 1991, when it was applied to populations at district level.16
Both experiences illustrate the highly political nature of resource allocation. The Swedish model ran into some difficulties when quite large shifts had to be achieved in the switch from the interim model in 1995 to the full model in 1996, particularly as the overall funding per inhabitant was falling over the same period. Although full implementation was agreed for 1996, a one-off compensation, as mentioned above, was given to the authority that stood to lose the most. Agreement on full implementation for 1997 was politically easier, as the shifts in funding were not as great.
In 1995 the York model was not implemented in full in Britain because of the government's nervousness over the size and direction of the implied shifts in resources, generally from suburban towards poorer areas. Identifying two separate models (table 1) allowed room for subsequent manoeuvre. There are even suggestions now that the market forces factor, introduced into the British formula by the Department of Health, is seriously undermining the model's attempt to allocate resources according to need.26
This illustrates the need to ask continually whether the policy as implemented is achieving its original objectives of equitable resource allocation.
Effects of cost containment
Finally, both approaches illustrate the complications of trying to devise and implement an equitable formula in a time of cost containment, when any redistribution of resources is much more painful. Some commentators suggest that the strain imposed by the prolonged underfunding of the British NHS in the 1980s was a key factor in the decision to overhaul the original formula created in the late 1970s.13 The drastic cuts that have had to take place in Sweden in the 1990s with the economic recession mean that the effects of resource allocation are not easy to disentangle from the effects of cutbacks.
Yet it is at just such times that efforts need to intensify. The joint effects of cutbacks and market-style reforms could be especially damaging to access to healthcare for the sections of the population in greatest need, as in a more competitive environment resources tend to flow to more prosperous areas and groups. It is important that the quest for equitable methods of resource allocation continues and is taken up by the growing number of other countries facing a similar challenge.
Funding: Stockholm County Council funded the development of the resource allocation model and its annual update.
Conflict of interest: None.