Study setting and design
We performed a retrospective study based on the Universal Health Insurance’s claims database (Système national des données de santé-SNDS) [10].
The Universal Health Insurance (UHI) fund covers more than 90% of the French population. The French health care system consists of primary care (mainly private practice) and of hospital care (mainly public sector) and has a cost sharing policy. The UHI fund reimburses physician private practice on the basis of a national fee schedule with reimbursement rates ranging from 30 to 100% of the statutory tariff for each type of procedure. These tariffs are set by national agreements among physicians’ trade unions and the UHI fund. For emergency department visits, tariffs and out-of-pocket payments are the same in all types of healthcare facilities. Personal health expenditure is mostly financed by UHI fund (79%). The remaining is financed by public or private complementary health insurance (14%) and out-of-pocket payments (7%) [11]. The complementary universal health coverage (CMU-C) is a public complementary health insurance for the poorest part of the French population (8%). Foreigners without any legal status can benefit from a specific public health insurance plan (Aide Médical d’Etat - AME).
This SNDS administrative database includes all outpatient visits (in private practice or healthcare facilities), inpatient admissions, medical procedures, medications, imaging, that are partially or fully covered under the universal health insurance fund. It contains also patient’s characteristics such as age, sex and municipality of residence. Patients presenting with one of 30 specific chronic long-term conditions (among which diabetes, coronary artery disease, heart or lung failure, psychiatric conditions, cancer, severe stroke, HIV, tuberculosis) are supported by a specific comprehensive coverage system for all related care. Each patient affected by one of these diseases needs physician certificates and administrative approval to benefit from this system. There are no complementary or out-of-pocket fees associated to these specific conditions. Administrative data concerning these conditions are also reported (Affection Longue Durée – ALD).
We included all patients who visited at least once any ED of the Ile-de-France region (IdF) between January 1st to December 31st, 2015. We excluded all visits for obstetric emergencies and childbirths. The IdF area is a region composed of eight départements, including the city of Paris. It has 87 general and 35 pediatric EDs.
The clinical severity of each ED visit was assessed according to the following classification:
-
need of a medical consultation only (level 1);
-
need of a technical procedure (biology test, imaging exam) (level 2)
-
need of a specialist opinion but without hospital admission (level 3);
-
resulting in hospital admission (level 4).
Definitions
There is no consensus in the literature on the definition of FU which ranges from ≥2 to ≥20 visits per year [4]. For this study, the definition of FU was based on the natural break in the distribution of ED visits in the studied population which allowed us to identify a cutoff of ≥3 visits per patient per year in any ED. This definition is also the most common one reported in the literature. Three subtypes of FUs were defined: low-FUs: [3,4,5,6] visits/year; high FUs: [7,8,9,10,11,12,13,14,15,16,17,18,19] visits/year; and very high FUs ≥ 20 visits/year [8].
Geographical units
We assessed a territorial division of the IdF region which is divided in 8 administrative departments. We identified 232 geographical units (GUs). GUs were defined according to the National Institute of Statistics and Economic Studies (INSEE) methodology: For the department of Paris, GUs correspond to districts (n = 20); For the three departments closest to Paris, GUs correspond to municipalities (n = 123); For the four peripheral and larger departments, GUs correspond to either municipalities (n = 10) or townships (n = 79) according to the size of the population. INSEE has developed this geographical subdivision to make the statistical data of the smallest municipalities more reliable. To characterize each GU, we used thirty descriptive variables of the demographic and socio-economic status of its residential population: age distribution, income level, jobs typology, education level, family composition, densities of healthcare professionals and the Human Development Index-2 (HDI2). The HDI-2 is an index which takes into account the three dimensions of the Human Development Index (health, education, standard of living) adapted to the French situation and available for each GU [12, 13].
Endpoints
The first objective of this study was to describe FUs in the specific IdF area. Our first endpoint is the FUs rate in the IdF region and at residential territory level. The secondary objectives of this study were to identify and characterize in terms of socio-demographics factors the territories the most associated to high FUs rates. Our secondary endpoint was to assess correlations between socio-demographics and economic characteristics of the geographic units and the FUs rate and to identify a typology of territories.
Statistical analysis
We defined the ED visiting rate by assessing the total number of ED visits per 1000 residents for each GU. All ED visits were recorded and accounted for even when the visit took place in a different GU than the patient’s residential GU. Then, the FU’s rate was assessed based on the number of FUs residents in a GU reported to the population of all GU residents that had visited at least once any ED in the IdF region.
We assessed correlations with the FU rate of 151 available demographic and socio-economic descriptive variables of all geographical units: age distribution, income level, jobs typology, education level, family composition, densities of healthcare professionals and the HDI-2. Among these, we identified 30 descriptive variables that were correlated to FU’s rate. Fifteen active interest variables were favored for principal component analysis (PCA) [14]. We moved from 15 variables to four main components while keeping more than 92% of the information. We performed a second step analysis by using a hierarchical ascending classification from these four main components. This method allowed the identification and differentiation of three different classes of homogeneous GUs with similar characteristics.
Data management and statistical analyses were performed with Excel® software, Microsoft Office Professional Plus 2010 for Windows® version; SAS Enterprise Guide®, version 7.13 of SAS System for Windows®; and IBM SPSS®, Statistics 20 version for Windows®.
Authorization
This study is based on public data extracted from the SNDS. Access to the data was granted to the authors as authorized personnel of the IdF Health Regional Agency. Conditions of access to the data are described in the French Decree No 2016–1871 of December 26th, 2016.