|
Modelling the Use of Hospital Services as a Function of Needs
and Supply in Italy for Capitation Purpose
Alessio Petrelli, Epidemiology Unit, Piedmont Region, Turin,
Italy, Roberta Picariello, Epidemiology Unit, Piedmont Region,
Turin, Italy, and Giuseppe Costa, University of Turin, Turin,
Italy
In economically developed countries, there has been an
increasing interest in the methods used for funding health care,
attempting to attain both equity and efficiency. One approach is
capitation, which is used in most countries whose health system
is based on taxation. Capitation entails determining weights
that represent health care needs and which are applied to
population subgroups to determine the funds to be assigned to
individual residents. Different methods and levels of refinement
for capitation have been used worldwide. In many countries, the
funds to be distributed are established based on the
population’s age structure, whereas in countries with greater
experience (e.g., England and Scandinavian countries), more
sophisticated formulas which take into account deprivation are
used [1].
In Italy, the health system is organised regionally (the
country is divided into 20 Regions), and most of the funds for
health care are provided by local taxes. The individual Regions
are responsible for choosing the specific method that the
Regional Health Authority will adopt for funding. In some
Regions, only historical expenditure is taken into account,
whereas others use basic capitation based on the population’s
age. The latter method is used in Piedmont (a region of
northwest Italy with approximately 4,000,000 inhabitants). In
this Region, we evaluated a method of health care funding by
modeling the use of health services as a function of factors of
direct and indirect needs and of supply indicators. The model
was greatly inspired by the model used in the United Kingdom and
is based on the consideration that the criteria used for
equitable funding should mainly consider the health status of
the population yet that the available data on morbidity are
incomplete both geographically and temporally. As a consequence
of the incompleteness of these data, in many countries health
care is funded on the basis of the use of health services.
In Piedmont, mortality and hospitalisation follow different
patterns, as shown in Figures 1 and 2. The distribution of
hospitalisation can in large part be explained by variations in the
access to health care. The western half of Piedmont is a mountainous
region characterised by isolated valleys, which create obstacles to
accessing health care, especially emergency care. In funding health
care based on the use of health services, there is the risk of also
funding inappropriate care; for this reason, “risk adjustment” for
taking into account health care needs is of particular importance.
Measuring needs is fundamental; in that capitation is the
expenditure of an individual for his or her needs. However,
morbidity data are insufficient, and a set of indicators of direct
and indirect needs must be used. The definition adopted to determine
which needs factors are important should include all of the factors
that affect the use of health services in a statistically
significant manner [2]. Although risk adjustment allows unsatisfied
needs, proportion of needs that do not meet demands, to be taken
into account, it also includes the effect of the “illegitimate”
needs. Variations in efficiency levels, political choices, and all
of the supply effects could be considered as illegitimate factors,
yet only some of the variability due to supply effects is
measurable. Given that the availability of resources is more of a
political issue, we focused on the factors to be considered in risk
adjustment and the weights to attribute to factors. The main
objective of the statistical model was that of weighing legitimate
factors of needs after controlling for illegitimate factors of
needs, concentrating on the variation generated by legitimate
factors more than on the evaluation of the total amount of
variability explained by the model. In the present work, we focus
on the critical issues that need to be addressed, as opposed to the
actual results of the model. Since the information from the
population census database at individual level was insufficient, we
conducted an ecologic study based on municipality data. Of the
municipalities in Piedmont, 80% have fewer than 3,000 inhabitants,
whereas 5% have more than 10,000 inhabitants. The Region’s capital
is Turin (population of approximately 900,000 inhabitants). Because
a historical population register has been active in Turin since 1971
(collecting demographic information on all residents), we used
neighbourhood-level measures for Turin to reduce ecological bias.
 We
integrated data from several information systems. In particular, the
Hospital Discharge Database (data from 2003) was used to measure the
use of hospital services, and the Regional Mortality Register
(2001), which includes data on the number of deaths at the
municipality level, was used to determine direct needs. The 1991
population census and the National Archive of Immigration (2001)
were used for indirect needs. As sources for the supply indicator,
we used the Regional Hospital Register (2001) and a regional
database of distances, which provides information on distances in
minutes between individual municipalities in Piedmont (travel by car
in normal traffic conditions). We used raw hospitalisation rates as
a measure of outcome, mortality rates as a measure of direct needs,
and age and a set of socioeconomic data as measures of indirect
needs. The socioeconomic parameters (i.e., percentage of the
population represented by immigrants, percentage with a low
educational level, percentage with a manual job, average number of
inhabitants per room, and percentage unemployed) were used because
the characteristics on which it is based are proxies of the
independent dimensions of the socioeconomic characteristics of a
given population. Moreover, expenditure not related to food
consumption was used as a proxy of mean income, and population
density (i.e., number of inhabitants per km2) was used as an
additional possible determinant of the use of health services. The
number of hospital beds and the distance from the nearest
municipality with a hospital were used as a measure of supply. We
first evaluated whether the assumption of independence among the
socioeconomic dimensions was tenable, fitting standard regression
models for testing multicollinearity; we then used hierarchical
regression models, considering the municipality as level 1 units and
the Regional Health Authorities as level 2 units. We used
hierarchical models to accurately estimate the uncertainty of
clustered data. All covariates were normalised around zero. All of
the values of the variance inflation indicator were much lower than
10, indicating that there were few problems with multicollinearity
(Table 1), which is consistent with the role of socioeconomic
indicators in explaining differences in the dimensions of
inequalities in health and in the use of health services. Table 2
shows the results produced by the hierarchical model. The percentage
of the population aged 65-84 years (coeff. 0.25 p < 0.05) and the
percentage with a manual job (coeff. 0.06 p < 0.05) were
statistically significant, indicating excesses in the use of
hospitalisation in these population subgroups. Of the illegitimate
factors, distance was inversely associated with hospital use (coeff,
-0.159, p < 0.05).
This is the first study in Italy to have estimated weights for the
purposes of capitation using socioeconomic factors as indirect
measures of need. The significance of the factors of indirect needs
suggests that deprivation plays a role in the estimate of needs,
especially for the occupational dimension represented by the
percentage of the population with a manual job, which contributes to
explaining the demand for hospital care. However, in interpreting
the results of this study, several limitations need to be mentioned.
The analyses were based on data aggregated at the municipality
level, which, as known, can produce an ecological bias. In Italy,
there are approximately 8100 municipalities distributed in 20
regions, and 15% of these municipalities are located in Piedmont.
Although these municipalities have a very low average population,
there also exist large urban centres (e.g., Turin and its suburbs
and the other capitals of the Region’s provinces). Thus, in addition
to the bias related to the aggregation of data, the heterogeneity of
the size of the individual municipalities must be taken into
account. Studies that have investigated the intensity of the
ecological bias in the relationship between socioeconomic factors
and health [3, 4] at diverse levels of geographic aggregation have
revealed that the regression coefficients are higher in the
aggregated analyses, compared to those in the individual analyses,
and that they remain stable with varying geographic granularity,
whereas the effect of geographic heterogeneity in the estimates is
not clear. In Piedmont, to overcome this problem, the possibility of
increasing the level of granularity in the census data, beginning
with the 2001 census data, is being evaluated. Another possible
limitation concerns the delays in the publication of census data,
which did not allow us to use the most recent census (2001).
Nonetheless, we assume that only slight changes have occurred in the
association between the use of hospital services and socioeconomic
levels. It would be useful to integrate socioeconomic indicators
with indicators of social networks, which are more sensitive to
problems concerning the access to health services related to
conditions of isolation. Finally, the problems of a statistical
nature should be evaluated more in depth, in particular: 1) the
presence and extent of endogeneity, possibly using models based on
linear equation systems, and 2) the presence of random slopes, which
could possibly be explained in terms of the conceptual suitability
of the model used. In particular, the mechanisms underlying the
transformation of needs into demand are in no way taken into account
in the model. Indicators capable of addressing the propensity
towards the consumption of services, which could depend on cultural
or environmental factors not explained by the indicators, could be
useful in measuring the phenomenon. Moreover, the use of
coefficients for implementing mechanisms for the funding of services
should take into account, although outside of the statistical model,
the diverse costs related to the production and distribution of
services, which are difficult to measure. Differences in the costs
of managing emergencies due to the terrain of a given area or in the
costs of distributing certain services (cleaning, cafeteria, etc.)
could be of importance. Another important aspect is the role that
the supply plays in the model. The strong effect of supply let
easily imagine a consistent gap between observed and estimated
funds. It is unlikely that the political level can decide to use a
model that does not consider the concentration of, and distance from
health services given this strong difference. The effects of
political choices able to make possible a decrease of supply could
act over medium-long periods, whereas capitation seems to be more
suitable for the funding of current expenses. More general political
problems are also not negligible, in particular, those related to
the introduction of a rather technical method in decision-making
processes. Finally, it is necessary to better evaluate the
implications of an approach that aim to reach an equitable objective
under the hypothesis that equity in access is equivalent to equity
in health outcome REFERENCES
- Rice N, Smith P. Approaches to capitation and risk adjustment
in health care: an international survey. Centre for health
economics, University of York, 1999.
- Rice N, Smith P. Capitation and risk adjustment in
health care financing: an international progress report. The
Milbank 2001;79:81-113.
- Soobader M, LeClere FB, Hadden W, Maury B. Using aggregate
geographic data to proxy individual socio-economic status: does
size matter? Am J Pub Health 2001;91:632-6.
- Geronimus AT, Bound J. Use of census-based aggregate variables
to proxy for socio-economic group: evidence from national samples.
Am J Epidemiol 1998;148:475-86.
|
