Sociodemographic and morbidity indicators of need in relation to the use of community health services: observational study

Introduction

The target allocation of funds to health authorities used to be determined by formulas that adjusted raw population data to allow for three factors—age, need (as measured by a variety of socio-demographic indicators), and market forces (to allow for differences in health authority costs). These formulas, which were derived from work undertaken at the University of York, applied separate adjustments to those portions of health authority budgets that covered expenditure on general and acute services and psychiatric services.1 2 3 When we undertook this study, however, no weighting for special needs was being applied to expenditure on community services. This had caused concern that the ability of the York recommendations to redistribute budgets had been lessened.4 The second report of the House of Commons health select committee considered the situation unsatisfactory and called for a weighting to be applied with effect from April 1997.5 We report work that examines the evidence for such weighting.

Methods

Ideally, we would have liked to have replicated, for community services, the study based on national data undertaken at the University of York. Unfortunately, no national data on community health services comparable to the hospital episode statistics currently exist. We were, however, able to obtain information from seven provider trusts, who supplied information on six programmes of care (not all trusts gave information on all programmes). Table 1 shows these programmes. The trusts served a mix of urban and rural areas. Table 2 gives the social demography of the areas served by these trusts and comparable figures for England. Small differences between the trusts and national averages exist, but none was significant. Nor was there any consistent pattern to suggest that the trust samples were more or less disadvantaged than England generally.

For each community health programme we determined an age adjusted use for each enumeration district in our sample and used a set of plausible sociodemographic variables to explain variations from this expected use. The postcode addresses of the patients were assigned to enumeration districts by using the methods suggested by Majeed et al.6 This established a link with sociodemographic data from the 1991 census. Since enumeration districts are too small for their standardised mortality ratios to be reliable, we attributed the standardised mortality of the relevant ward to its constituent enumeration districts. We used multiple regression analysis (from release 6.1 of SPSS for Windows) with dummy variables to capture any effects specific to individual trusts (for example, differences in the availability of resources or attitudes toward the use of services). Rather than risk spurious correlations by including all possible population descriptors, we restricted our dataset to those factors shown by a review of the medical literature to be probable determinants for community services. A difficulty arose because trusts do not have clearly defined boundaries. Accordingly, we decided to exclude data from all enumeration districts in those wards that seemed to lie on the periphery of trust catchment areas.

Results

We examined the correlations between our potential explanatory variables and found that many were highly intercorrelated. In such a situation we would have expected our model to be “fragile”—those variables selected in the model and the magnitudes of their regression coefficients might have differed substantially in response to minor changes in the dataset. This does not necessarily imply, however, that predictions from the model will be incorrect. Models for the individual care programmes (table 3) were parsimonious (with no single formula using more than four explanatory variables) and highly significant. All regression coefficients had the expected sign, with greater use of services associated with higher levels of deprivation or morbidity.

Discussion

Our general approach is similar to that used by the researchers at the University of York.1 2 3 Consequently, it is subject to some of the same criticisms. As the York researchers observed, “No study using a methodology based on utilization can capture variations in health care needs that are not reflected in utilization.” Our technique essentially abstracts the systematic component of historical decisions. If historical decisions have been systematically incorrect, as might occur if the “inverse care law” applies to community health services, our conclusions will be too.7 The assumption implicit in our method is that while individual decisions in the past may have been wrong, the use of community healthcare services has generally reflected the needs of patients.

An alternative method, which does not rely on this assumption and is therefore robust in the face of problems that might arise from the effects of the inverse care law, exists. This would be to identify the important diseases treated in the community, then to determine the relation between those diseases and various proxies for need. With further information about the cost of treatments we could synthesise resource needs based on the relation between illness and social characteristics rather than rely on the relation between the historical use of services and social factors. We explored the feasibility of adopting such an approach but found that some important items of data were not available. Information on the case load of community services in relation to disease was poor and that on diseases treated in the community (as best we could identify them) and social factors was incomplete. A large scale survey would be necessary to collect this information—we had neither the time nor the resources.

A further fundamental problem, for which no simple remedy exists, arises from the “ecological fallacy.” Our method assumes that the social and other characteristics of the enumeration districts in our study have influenced the use of services. Users of community services may not be typical members of the communities in which they live, but our decision to analyse data in relation to the small enumeration districts should have minimised the dangers of adopting this general approach.

Our results suggest that an adjustment to the community element of the weighted capitation formula should be made to allow for the socio-demographic and morbidity characteristics of the populations that health authorities serve. Such an adjustment might be achieved simply by extending the existing indices to that area of expenditure. Our calculations suggest, however, that the existing psychiatric index differs from the community mental illness index we have estimated in the extent to which they redistribute resources. It also imposes the restrictive assumption that the ranking of areas by needs for community health services is the same as that for inpatient services. Perhaps for these reasons, our results have now been incorporated into the Department of Health's formula for the allocation of resources to health authorities.