Clinical Trial Summary
Many studies have shown that deprived patients consumed more healthcare resources than
non-deprived patients, in particular in terms of increased length of stay (LOS) and
readmission rates, which has an impact on hospital efficiency and the healthcare system as a
whole. There are many types of indicators available to assess deprivation in a hospital
setting and French decision makers are currently using reliance on public aids to allocate
additional funding to hospitals, based on the percentage of deprived patients they admit.
However there are limits to this method: it only assesses one dimension of deprivation, the
target population often does not know about the existence of those aids, and they have a
clear threshold effect. An alternative solution is to use ecological deprivation indices
which are obtained by aggregating different variables measured at a specific time and place,
i.e. the patient's place of residence at the time of care. One such index, the FDep , was
developed specifically in France, although others such as the Carstairs index and the
European deprivation index also exist.
The primary objective of this study is to study the association between deprivation, measured
by the FDep, and hospital care efficiency in paediatric and neonatology patients, measured by
the difference between patient LOS and the national average LOS of their diagnosis-related
group, DRG). The secondary objectives are to carry out a budget impact analysis on the impact
of deprivation for hospitals with a paediatric or neonatology ward, to study the association
between deprivation and readmission at 15 days, to study the relation between FDep and the
currently used deprivation indicators, and to assess the added value of the FDep compared to
those indicators and whether or not it should be used in routine practice.
In order to do so, an exhaustive retrospective study using the French hospital claims
database will be carried out for the years 2012-2014. Deprivation indices will be calculated
based on patients' postcode. The primary endpoint will be calculated using the national LOS
present in the French national cost study. Similarly, the budget impact will look at the
difference between production costs derived from the national cost study after adjusting for
LOS and the statutory health insurance's tariffs, which will allow us to assess whether a
hospital stay is associated with a gain, a loss or is budget-neutral for the hospital.
Readmissions at 15 days will be identified through record linkage.
Descriptive analyses will summarise both hospital and patient characteristics. Uni- and
bivariate analyses will be carried out by focusing of the variables of interest (e.g. average
deprivation index by legal status of the hospital, mean LOS depending on the number of
paediatric beds etc.). The deprivation index will be divided into quantiles as is the norm
and the endpoints will be assessed for each of those quantiles. An ANOVA (or a Kruskal-Wallis
test if the ANOVA hypotheses are not met) will test whether the results differ between each
quantile. For readmission rates, a Chi² test will be performed.
In order to study the association between deprivation and the endpoints, the investigators
will model each endpoint using as the main explanatory variable the deprivation index. Three
main types of explanatory variables will be added to the model: patient characteristics (age,
sex, severity level etc.), hospital characteristics (legal status, size, number of full-time
equivalent etc.) and environment/context characteristics (number of paediatricians for 1,000
inhabitants, rural vs. urban area etc.).
In order to assess the added benefit of using the deprivation index vs. the current
indicators, a sub-cohort will be constructed in Paris teaching hospitals (AP-HP) as
unfortunately, whether the patient receives public aids is not present in the hospital claim
database but is available only at the local level. The investigators will look at the
distribution of patients with public aids in each quantile of the deprivation index and run
the previous models using the two types of indicators one after the other and comparing the
statistical performance of each pair of models.
