Date : JUIN 15

Page de l'article : p.51-57

Journaliste : G. Moulis / M.
Pays : France Lapeyre-Mestre / A. Palmaro /
Priodicit : Mensuel G. Pugnet / J.-L Montastruc /
L. Sailler

Page 1/7

Review

French health insurance databases: What interest for medical CrossMark

research?1^
Les bases de donnes de l'assurance maladie franaise : quel intrt
pour la recherche mdicale ?
G. Moulis a '*' b ' c , M. Lapeyre-Mestreb'c'd, A. Palmaro b ' c , G. Pugnet a ' b ' c ,
J.-L Montastruc b ' c ' d ' e , L Saillera'b'c
a
Service de mdecine interne, salle Le Tallec, CHU Toulouse Purpan, place du Dr-Baylac, ISA 40031,31 059 Toulouse cedex 9, France
b
UMR 027, universit de Toulouse III, 31000 Toulouse, France
c
Inserm UMR 1027, quipe 6 pharmacopidmiologie, valuation de l'utilisation et du risque mdicamenteux, 31000 Toulouse, France
A
Service de pharmacologie mdicale et clinique, facult de mdecine, 31000 Toulouse, France
e
Centre Midi-Pyrnes de pharmacovigilance, de pharmacopidmiologie et d'informations sur le mdicament, CHU de Toulouse, 31059 Toulouse, France

A R T I C L E I N F O A B S T R A C T

Article history French health insurance databases are organized smce 2003 into a huge digital data warehouse, the
Available online 26 December 2014 Systme national d'information mter-rgime de /'assurance maladie (SNIIR-AM) It covers the entire French
population (65 million mhabitants). In order to facilitate studies on more frquent conditions, a random
Keywords sample of I /97th of national health system beneficianes has been built smce 2005, called the chantillon
Health Insurance gnraliste des bnficiaires (EGB). The aim of this article is to descnbe the main charactenstics of the
Mdico-administrative databases SNIIR-AM and the EGB, to dtail their accessibilite accordmg to French law, and to present their strengths
France
and lumts. It is illustrated with the most rcent studies conducted in these databases. These databases
Systeme national d'information mter-regime
de l'assurance maladie
include dmographie, out-hospital reimbursement (includmg drug dispensmg), medical (costly long-
chantillon gnraliste des bnficiaires term diseases, occupational diseases, sick-leaves ), and m-hospital data All these data are prospective^
recorded, individualized, made anonymous and hnkable Consequently, the SNIIR-AM is a very useful data
source for epidemiology!, pharmacoepidemiological and health economies studies, particularly for rare
diseases. The EGB is appropnate for long-term research on more frquent diseases.
O 2014 Socit nationale franaise de mdecine interne (SNFMI). Published by Elsevier Masson SAS.
All nghts reserved.

R E S U M E

Mots cles
Assurance maladie
Base de donnes mdico-administrative
France
Systeme national d'information
mter-rgime de l'assurance maladie
chantillon gnraliste des bnficiaires
Les bases de donnes de l'assurance maladie sont collectes depuis 2003 dans un vaste entrept
numrique, le Systme national d'information mter-rgime de l'assurance maladie (SNIIR-AM). La rsul-
tante en est une des plus grandes bases mdico-administratives au monde, couvrant 65 millions de
personnes Afin de faciliter l'tude de cohortes de patients atteints de maladies plus frquentes, un
chantillon au I /97e des assurs l'assurance maladie a t constitu depuis 2005 l'chantillon gnral-
iste des bnficiaires (EGB). L'objectif de cette mise au point est de prsenter les grandes lignes de
l'architecture du SNIIR-AM et de l'EGB, leurs modalits d'accs, leurs intrts et leurs limites. Leur
potentiel en recherche mdicale est illustr par les publications les plus rcentes. Ces bases de donnes
contiennent des donnes dmographiques, les donnes de remboursements des prestations ambulatoires
(dont les dlivrances de mdicaments), les donnes mdicales des rgimes de l'assurance maladie (affec-
tions de longue dure, maladies professionnelles, arrts de travail ) et les donnes hospitalires issues
du programme de mdicalisation des systmes d'information. Toutes ces donnes sont individuelles,

* The French version of this article is available online (see Appendix A at the end of the text)
* Corresponding author
E-mail address guillaume moulis@univ-tlse3 fr (G Moulis)

Tous droits rservs l'diteur SANTE2 6564384400503


Pays : France
Priodicit : Mensuel
prospectivement recueillies, anonymises et chanables Tout cela fait du SNIIR-AM une source de donnes
trs intressante pour la recherche pidmiologique, pharmacopidmiologique et en conomie de la
sant, particulirement pour les maladies rares L'EGB est particulirement utile l'tude des maladies
plus frquentes et sur le long terme
2014 Socit nationale franaise de mdecine interne (SNFMI). Publi par Elsevier Masson SAS Tous
As in North American and other European countnes. French
public authonties want to develop the use of national mdico-
administrative databases for research. For that purpose, their
accessibihty to researchers is regularly improved [1-3]. This arti-
cle is focused on the use of the French national health insurance
system databases for medical research.
The French universal health insurance system manages, through
public antilles, all reimbursements of healthcare to all people
affihated to a health insurance scheme in France Accounts manage-
ment requires the generation of centralized databases, containing
also administrative data. The resuit is one of the largest global
database of medical and administrative data in the general pop-
ulation, covenng nearly 65 million persons, fai ahead of the
UK's Clmical Practice Research Datalmk (approximately five mil-
lion persons), the Quebec Health Insurance Corporation (Rgie de
l'assurance maladie du Qubec, approximately four million per-
sons) oi these of northern European countnes [I ]. Healthcare data
are collected prospective^ at the local level, and then transmit-
ted to the National Health Insurance Fund for Salaned Workers
(Caisse nationale de l'assurance maladie des travailleurs salaris
[CNAMTS]) that hosts all the data in a huge digital warehouse
called the National Health Insurance Inter-Scheme Information
System (Systme national d'information mter-rgime de l'assurance
maladie [SNIIR-AM]) [1,4]. All SNIIR-AM data are hnkable: it is
possible to identify and to link all data corresponding to a given
beneficiary among the dozen of tables contamed in the SNIIR-
AM.
Due to its width and to the individual, prospective and com-
plte data recording for all beneficianes, the SNIIR-AM is a very
useful data source for medical research in the field of epidemiol-
ogy, pharmacoepidermology and health economies of rare diseases.
To facihtate the building of patient cohorts for the most common
diseases, a sample of l/97th of the SNIIR-AM bas been established
smce 2005 in order to provide a 20-year follow-up of beneficianes
This sample is named the permanent beneficianes sample (chan-
tillon gnraliste de bnficiaires [EGB]) [4,5].
The ann of this article is to descnbe the main charactenstics
of the SNIIR-AM and the EGB aichitectuies, to piesent the condi-
tions for accessing to them and their main strengths and limitations
when used for medical research. The potential of these databases
in medical research is illustrated by the most rcent publications.
I. Simplifie! architecture of the SNIIR-AM and the EGB
1.1 The SNIIR-AM
The SNIIR-AM was created in 2003 following the 1999 social
secunty fundmg law [6]. Its objectives are to improve the qual-
ity of healthcare, contribute to a hetier management of the Heallh
Insurance Syslem and of public heallh polices, and inform heallh
praclilioners as regards Iheir activity, accounts and prescnplions
[7]. Initially consistmg of only CNAMTS dala (86% of insured
palienls), il was gradually ennched in 2009 by lhc dala of all
insured patients by linkage with information of the programme for
the medicalization of information systems (programme de mdi-
calisation des systmes d'information [PMSI]) of public and private
Tous droits rservs l'diteur
Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Lapeyre-Mestre / A. Palmaro /
G. Pugnet / J.-L Montastruc /
L. Sailler
Page 2/7
droits rserves
hospitals fi om 2007 to 2010, and by linkage with dates of death
(smce2009) [I].
Its general architecture and main recorded data are descnbed in
Fig I.
The inter-scheme consumption data (donnes de consomma-
tion mter-rgimes [DCIR]) set includes all outpatient reimbursed
health expenditures. These data are transmitted in real time by the
CNAMTS and the various special insurance schemes.
This DCIR database is also linked with data from the health insur-
ance medical service that includes information about long-term
diseases [LTD), occupational accidents, and occupational diseases.
PMSI data stern from all private or public hospitals and are pro-
vided to the CNAMTS for linkage to the SNIIR-AM. Information
includes the hospital identifier, the dates of start and end of stays,
the diagnostic codes (one primary diagnosis, one related diagno-
sis and up to 30 associated diagnoses coded with the International
Classification of Diseases, version ICD-10 [8]), the diagnosis-related
group (groupe homogne de malades [GHM]) codmg, the proce-
dures (Classification commune des actes mdicaux [CCAMJ codmg),
the costly drugs dispensed, whether the stays occurred in a special
unit (resuscitation, intensive care, neonatology or palliative care),
and a score of seventy for stays in intensive care umts (Simphfied
Acute Physiology Score ll). A database exists for Mdiane, Surgery
and Obstetncs hospitalisations (PMSI MCG for mdecine chirurgie
obsttrique), one for psychiatry (PMSI PSY for psychiatrie), one for
home hospitalisations (PMSI HAD for hospitalisaiton domicile), and
one for after-care and rehabilitation (PMSI SSR for soins de suite et
de radaptation) PMSI data are updated every three months [1,4].
Demographics (age, gender, place of residence -mumcipality
code, advantage of the Universal Health Coverage, affihate scheme)
are available in the DCIR and PMSI databases.
Data of a given individual are hnked through a unique identifi-
cation number named numro d'identification au rpertoire (NIR),
which is a unique identifier for each insured person. However,
the NIR is not accessible in the SNIIR-AM in order to prserve the
identity of patients. It is made anonymous by a complex two level
procedure called/cmcfion d'occultation des informations nominatives
(FOIN), winch is completely irrversible This makes it difficult to
match climcal cohort data with SNIIR-AM data (Sec Section 2) [I].
The depth of the SNIIR-AM is three years plus the current year.
PMSI data kept by the Technical Agency for Hospitalisation Infor-
mation (Agence technique d'information sur l'hospitalisation [ATIH])
have a shding depth of IO years plus the current year. The law allows
conservation of successive SNIIR-AM extractions corresponding to
lOyears of follow-up [7].
12. The ECS
The EGB lias been established in 2005. It is a sample of l/97th
of insured persons in France, corresponding to 660,000 persons
[4,5]. Il is built by a random selection of NIRs repiesentative of
the French population by gender and by five-year incrments [4]
The population of the EGB includes insured persons, whether they
are receivmg healthcare or not In contrast, only individual receiv-
ing healthcare are present in the SNIIR-AM. All health insuiance
schemes are included, except some special insurance schemes,
SANTE2 6564384400503
 Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Pays : France Lapeyre-Mestre / A. Palmaro /
Priodicit : Mensuel G. Pugnet / J.-L Montastruc /
L. Sailler

Page 3/7

Health care dispensing Civil status office

Departmental Health insurance Agencies Agences KegionatesaeSante INSEE

e l'Informmmn

\ i l'Hospitalisatia

Medical data
Costly long term diseases
Occupational accidents
Occupational diseases
Sick leaves
Dmographie data
Age, gender, place of
residency insurance scheme
benefrtfrom the Universal
Health Coverage
PMSI MCG, PSY, SSR, HAD Date of death
Entry and release dates
Principal, related and associated diagnoses
Procedures
Costly drugs
Special unit intensive care palliative care

Out-hospital reimbursements
Date person who prescribes and who dispenses the care
For drugs name, form, quantity dispensed

DOR datamart PMSI datamart

SNIIR-AM (hosted by the CNAMTS)

Fig I Simplifie! aicluteUure of the french National Health Insuiance Information System CNAMTS Caisse nationale de I assistance Maladie des (rural/leurs sa/anes (National
Health Insurance Fund for Salaned Workers) DCIR donnees de consommation inter regimes (Inter Scheme Consumption Data) HAD hospitalisation a domicile (Home Hos
pitalization) INSEE Institut national des statistiques et des etudes economiques (National Institute of Statistics and Economie Studies) MCG medecine chirurgie obstetnque
(Mdiane Surgery Obstetncs) PMSI programme de mdicalisation des systemes dinformation (Program for the Medicalization of Information Systems) PSY psychiatry,
SNIIR AM Systeme nanona/d m/ormanon!nrer-reimede/nssurancemafadie(NationalHealth Insurance Information System) SSR servicesdesuireerdereadaptcmon(after care
and rehabilitation)

whose data are progressively implemented Data available are
these of the SNIIR-AM They are also anonymous Individual data
are stored and accessible for 20 years When a patient dies emi-
grates or is no longer msured, ne or she is considered out of the
cohort his/her data are retamed but censored on the release date
Newly msured pei sons are added quarterly to the EGB to keep con-
stant the proportion of l/97th of the msured persons in France
[4]
Table I compares the SNIIR-AM with EGB Because of the depth
of the database and accessibihty, the EGB is a database of choice for
long-term cohorts of patients with common diseases
2. Conditions for accessing the SNIIR-AM and the ECB
Conditions for accessing the SNIIR-AM and the EGB are stnctly
regulated by law [7] The Health Data Institute (Institut des donnees
de sante [IDS]) ensures the consistency of mdico-administrative
databases, data quality and availabihty of databases for research
[9] Any access request to the SNIIR-AM or the EGB for a pioject
requires automation from the IDS Only non-profit research can
apply[10]
Access to the EGB requires a direct secured connection to a
devoted server, with no possibihty to import data As there is no
database creation, only authonzation from the IDS is required
Tous droits rservs l'diteur
Direct access to the SNIIR-AM is not possible Followmg autho-
nzation from the IDS and the National Data Protection Commission
(Commission nationale de l'informatique et des liberts [CNIL]),
CNAMTS engineers extract from the SNIR-AM the study popula-
tion accordmg to the inclusion and non-inclusion cnteria given to
them Data are supplied on a secure, electronic mdium
As noted above, SNIIR-AM and EGB data are irreversibly anony-
mous Collecting the patients' NIR in a chnical cohort (in order to
link cohort data with SNIIR-AM data) currently requires a State
Council (Conseil d'Etat) decree, making this an almost impossible
strategy in practice A second strategy is to identify the NIRs achiev-
ing a probabihstic matching It consists in recording in the chnical
cohort indirect identifymg data (date and place of birth, gender),
plus possibly directly identifymg data (name and surname) Then,
these data are seai cheri in the NIR i epertory in ot dei to ti y to iden-
tify the NIR that corresponds to a given patient However this later
strategy can lead to errors and non-completeness in NIR identifi
cation [I]
3. Advantages and limitations
The advantages and limitations of the SNIIR-AM and of the
EGB with existmg or planned solutions to the limitations are
SANTE2 6564384400503
 Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Pays : France Lapeyre-Mestre / A. Palmaro /
Priodicit : Mensuel G. Pugnet / J.-L Montastruc /
L. Sailler

Page 4/7

Table I
Companson of the Systeme national d information inter regime de I assurance maladie (SMUR AM) to the chantillon gnraliste des bnficiaires (EGB)

SNIIR AM ECB
Population All insurance schemes (smce 2009) Dynamic sample of I /97th ( n = 660 DOO) of msured persons in France
Only individual receivmg reimbursed healthcare (patients) except SI M and some rare insurance schemes (not necessanly ill)
Benefiting trom health care or not
Rareproblemoflossoffollow up
Database depth 3 yeais + thecurrentyear lOyears fat the PMSI Recording durmg 20 years (start in 2005)
Possibihty to stock data (bf IQ years
Data Out-hospital reimbursed healthcare Out hospital reimbursed health cares
ALD work accidents occupational diseases sick leaves ALD work accidents occupational diseases sick leaves
PMSI MCG PSY HAD SSR PMSI MCG only
Access Non profit research organization Non profit research organization
1DS authonzation IDS authorization
CNIL authorization Direct access by connection to ECB server
Extraction by CNAMTS engineers
Preferential use Studies on rare diseases Diseases with high prevalence
Studies needing rapid results
ALD affection de longue duree (long term disease) CNAMTS Caisse nationale de lassurance maladie des travailleurs salaries (National Health Insurance Fund for Salaned
Workers) CNIL Commission nationale de I informatique et des liberts (National Data Protection Commission) EGB chantillon gnraliste des bnficiaires (permanent
beneficianes sample) HAD hospitalisation a domicile (home hospitalization) IDS mstitutdesdonneesdesanre(HealthDatalnstitute) PMSI programme de mdicalisation des
systemes d information (Program for the Medicahzation of Information Systems) PSY psychiatry SLM sections locales mutualistes SMUR AM Systeme national d m/ormation
inter regime de I assurance maladie SSR soins de suite et de readaptation (after care and rehabilitation)

summanzed in Table 2 The main uses of these databases for med
ical research are hsted in Table 3
The mam limitation of these databases is their complexity
An extraction of the SNIIR-AM bas several dozen of tables, some
only used for accountancy, containmg hundreds of variables and
millions of hnes A SNIIR AM dictionary is available for users of
the health insurance databases As the EGB is dedicated to medi-
cal research, a 'simplified EGB' emerged in late 2013, simplifymg
the presentation of outpatient healthcares However, training on
the complex architecture of these bases is mandatory smce 2013

Table 2
Advantages and limitations of the SNIIR AM and the EGB

Advantages Limitations Rcent proposai for improvement

Architecture Overall sight of the patient s care
program
Financial data (health economies
studies)
Prospective recording of the data
Depth of the database 2005 2025
(EGB)
Insured persons All msured persons in France
healthcare consumption in a
real world practice
Out-hospital healthcare data Date and price of the health care
identity of the prsenter and of the
dispenser
Drugs date exact pharmaceutical
form number of dispensed umts
In-hospital data All hospital stays in public or
private hospitals in France
Up to 32 diagnosis codes
piocedures costly diugdispensmg
hospitalization in special care
departments (intensive care umts
neonatology )
Access Easy individual access to the EGB
server
CNAV Caisse natwnaled assurance vieillesse( National Fund for Agemg) ECB chantillon generale des bnficiaires (permanent beneficianes sample) HAD hospitalisation
a domicile (home hospitalization) NIR numero dinsertion au rpertoire (unique identification number for any msured adult) PMSI programme de mdicalisation des
systemes d mfounatwn (Program foi the Medicahzation of Information Systems) PSY psychiatry SLM sections locales mutualistes SNIIR AM Systeme national d information
inter regime de I assurance maladie SSR soins de suite et de readaptation (after-care and rehabilitation)
Complexity constant evolution
Time varyingdepth of3 years + current
year (SNIIR AM)
Complexity to identify healthcare for
children due to the possibility of
several nghts holder
No detailed socio professional data
Absence of the SLM in the EGB
No recording of the indications
No recording of the results of
paraclinical exammations
Drugs prescription data are not
lecoided over the countei drugs are
not recorded
Isolated diagnosis codes sometimes
not rehable
No detailed chnical data no drug
exposure data excepted costly
treatments
PMSI PSY HAD and SSR poorly
featured in the SNIIR AM and absent
fram the EGB
No data from medico social structures
housmg agemg people
No data for consultation if no
procedure is pei foi med
17 months on average to obtain
SNIIR AM extraction
No direct access to the MIR
Training by the CNAMTS and the IDS
(currently available)
Creation of the simplified EGB
(end 2013)
Increase of SNIIR AM depth and
possibility to stock conscutive
extractions
Creation of a national health identifier
Lmkage with CNAV data
Addition to SLM data currently
implemented in the EGB
Pharmacoepidemiologicalapproaches
for modehzation of drug exposure
Algonthms combinmg LTD and
hospitalization diagnosis codes
procedures and drugs
Improvement of PMSI PSY HAD and
SSR encodmg PMSI PSY HAD et SSR
currently implemented in the EGB
Addition of data fi om mdico-social
structures for agemg people (currently
tested)
Coding of consultations
Improvement of accessibilite to
SNIIR AM data
Draftmg by the State Council of a
deciee serving as general framework
for secure access to the NIR

Tous droits rservs l'diteur SANTE2 6564384400503


Pays : France
Priodicit : Mensuel
Table 3
Main uses of the French health insurance system databases for medical research
Field Main uses
Epidemiology Assessment of incidence and prevalence of
diseases
Assessment of geographical variations of
incidence and prevalence of
diseases description ot healthcare programs
(procedures exammations performed for a
given disease)
Assessment of the impact of guidelmes on
healthcare
Pharmacoepidemiology Assessment of vaccine coveiage
Quantification of drug exposure
Assessment of rare adverse drug events
Assessment of drug effectiveness in the
world life practice mcluding comparions not
carned out in trials before marketing
Assessment of the impact of health decision or
campaign on drug consumption
Health economies Assessment of healthcaie costs (procedees
exammations work stoppage sick leaves
tiansport )
Cost effecfiveness compensons of therapeutic
or prevention strategies
The EGB and the SNIIR-AM user groups meet regularly with offi-
ciais of the IDS and the CNAMTS to keep abreast of updates, and
commumcate on secure forums The EMOIS (valuation, manage-
ment, organisation, information, sant] group organizes meetings
and pubhshes reports to promote and improve research on the
SNIIR-AM and the EGB [ll] Working on these data also requires
sufficient knowledge of data management using the statistical soft-
ware SAS (SAS Institute, North Cambria, USA) used on the EGB
server Working on SNIIR-AM data extractions requires to have a
secured server with sufficient capacity
As previously suggested, a second general limitation is the
access to SNIIR-AM data extractions Obtaining the necessary
authonzations and data extraction by CNAMTS engineers requires
17 months on average [10].
3 I Advantages and limitations in epidemiology
The major advantage of these databases is the huge number of
patients they include and the prospective data collection There-
fore, large cohorts can be built in these databases to clanfy the
epidemiology of diseases at a nationwide level and takmg into
account geographical variations The SNIIR-AM is a very useful
data source to estimate the incidence and prevalence of rare dis-
eases, such as lupus or immune thrombocytopenie [12,13] For
instance, we could identify 3771 incident immune thrombocy-
topema patients in the SNIIR-AM during a two-year penod [13]
For companson, the most important cohort of immune thrombo-
cytopema patients ever built in mdico-administrative databases
was in the UK Clinical Practice Research Datahnk that identified
1145 incident adults during a 15 year penod The huge number of
patients followed in the SNIIR-AM allows the assessment of varia-
tions according to age, gender, seasons or areas [12,13]
A possible limitation in the use of mdico-administrative
databases in epidemiology is the vahdity of diagnosis codes Quality
control before data transmission to the CNAMTS and audits involves
mamly the treatment of aberrant or missing data [4] Moreover,
financial coding of healthcare is preferentially checked LTD attri-
butions are not exhaustive for a given disease for example, patients
who have already obfamed a LTD for diabetes mellitus and suffer-
mg fram myocardial infection do not systematically request for
a new LTD for myocardial mfarction [I ]. In the PMSI, the dtails of
Tous droits rservs l'diteur
Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Lapeyre-Mestre / A. Palmaro /
G. Pugnet / J.-L Montastruc /
L. Sailler
Page 5/7
diagnosis codes (pnmary, related and associated) should be studied
and not the GHMs, which form an accounting harmonization Lim-
itations in the accuracy of coding are inhrent to those of the
ICD-10 [8] Consequently, several French researchers and CNAMTS
engineers develop algonthms for the identification of prvalent
or incident patients that combine LTD and PMSI diagnosis codes,
medical procedures and drugs used as proxies (such as antidiabetic
or hypoglycaemic drugs to identify diabetic patients) Recently, an
approach combinmg diagnosis codes, endoscopie and surgical pro-
cedures demonstrated its efficacy in identifymg incident patients
with colonie cancer [14] Similarly, an estimation method of the
Charlson comorbidity score, an important covanable in epidemi-
ology, bas beer^validated in the SNIIR-AM [15] It uses similar
algonthms to identify comoibidities
A second limitation is the lack of detailed clinical data and of
results of paraclmical exammations Only events accompanying the
diagnosis or complicating the disease are encoded in the PMSI The
mability to collect the NIR prevents any reliable connection with
clinical data Rcent reports of the High Council of Public Health
(Haut Conseil de la Sant Publique) [I] and Bras-Loth report com-
missioned by the Minister of Social Affairs and Health [2] stress
the urgency of drafting a State Council decree serving as a general
framework for secured access to the NIR
Moreover, the PMSI database dees not collect the entire hospital
activites in the absence of procedure, consultations and emer-
gency department stays lasting less than 24 hours are not recorded
The PMSI PSY, SSR and HAD are still poorly featured in the SNIIR-AM
and absent fram the EGB
Patients are identified thanks to their NIR That makes studies
carried out in pdiatrie populations more complex Indeed, children
do not have their own rights as insured persons Consumption of
care is through their "rights holdei" (usually the legally guardian)
As a cniid may have several rights holders (beth parents, for exam-
ple) to facilitate access to care, the consumption of care of one
diiid can be linked in the SNIIR-AM with several identifiers A
national health identifier should soon be created to solve this prob-
lem|l]
Lastly, dmographie data are limited Only ail-cause mortalite
can now be evaluated in the SNIIR-AM as in a rcent follow-up
study in the general population after initial hospitalisation for
heart failure [16] The causes of death should be specified in the
SNIIR-AM in the near future by individual hnkage of the causes
of death from the epidemiology centre for causes of death (Centre
d'epidemwlogie des causes de deces [CpiDc]) of the National Insti-
tute of Health and Medical Research (Institut national de la sant et
de la recherche mdicale [Inserm]) [ I ] Lack of socio-professional
data could be completed in the future by hnkage of data with
the National Fund for Ageing (Caisse nationale de vieillesse) [I]. A
score of social deprivation [17] may also soon be added in the
EGB
3 2 Advantages and limitations in pharmacoepidemiology
The field of pharmacoepidemiology assesses the use, the effec-
tiveness and the safety of drugs in a real-world practice (after
marketing authonzation) Therefore, health insurance databases
are frequently used as a material for pharmacoepidemiological
studies They are the most appropriate tools to evaluate the immun-
isation coverage in the general population [18,19] or spcifie
populations We have estimated the exposure to pneumococcal
vaccine in adult patients with pnmary immune thrombocytopema
treated with ntuximab [20] as recommended by French guidelmes
[21] The EGB was also used to assess the relative frequency of pre-
scnbmgbiotherapies in rheumatoid arthritis [22] Health insurance
databases are tools of choice to assess compliance with the rec-
ommendations at the population level, as shown recently by the
SANTE2 6564384400503

Pays : France
Priodicit : Mensuel
exposure to recommended drugs after myocardial infarction [23],
However, they cannet be used to assess accurately the reasons for
non-adhrence to recommendations.
The depth of the EGB make possible a dynamic view of drug
exposure: our team was able to show the increasing number of
patients exposed to baclofen in rcent years for alcohol abuse
[24] and the impact of health authonties alerts about safety of
antipsychotics in elderly patients on the rate of prescription of psy-
chotrope drugs in this population [25]. Moreover, the SNIIR-AM
and the EGB allow the identification of consumer profiles [26],
Drug effectiveness in real-world practice can be assessed from
the SNIIR-AM or the EGB, by overall mortahty (soon, cause-specific
mortality), and senous events (m-hospital and LTD diagnosis codes,
procedures or drugs used as proxies). These databases are a pow-
erful material to compare the effkacy of two drugs with the same
indication, but which have not been compaied in clinical riials.
Moieover, they allow stratification of the analyses on subgioups
of patients excluded fi om clinical trials (elderly patients with
comorbidities, poly-medicated, etc.) due to the large number of
patients they mclude. Recently, a study conducted in the SNIIR-AM
and including 163,801 patients showed no superiority ofrosuvas-
tatin compared with simvastatm used for primary prevention in
reducing mortality and the occurrence of stroke and myocardial
infarction [27].
The SNIIR-AM and the EGB are particularly suitable bases for
post-marketing authonzation safety studies, in order to assess
the occurrence of rare adverse events [28-30]. Such studies have
previously led to decisions of public health as regards drug market-
ing. Of note, these safety studies in the SNIIR-AM are not limited
to drugs: a study assessing the impact of complications after
colonoscopy (such as perforation and gastrointestmal bleeding)
nas been recently published [31]. As regards immunopathology,
these databases are appropnate for the follow-up in the long-term
of cohorts of patients suffenng from rare diseases. Two studies
carried out in a Canadian database similar to the SNIIR-AM (the
Rgie de l'assurance maladie du Qubec database) assessed the nsk of
infection on corticosteroids in older rheumatoid arthrite patients
[32,33]. Our team is currently working in the SNIIR-AM with a
similar approach on the benefit-to-risk balance of the diffrent
therapeutic strategies used in primary immune thrombocytopema
(FAITH study, No.ENCePP 4574) [34]. Lastly, in-hospital and out-
hospital financial data are extremely detailed, and research in
the field of health economies is nsing in the SNIIR-AM and the
EGB.
However, some limitations should be kept in mind: as men-
tioned previously, the indications of treatments and of parachnical
examinations are not recorded. They are deducted from LTD and
hospital diagnosis codes, as well as from some procedures or drugs
used as proxies. Only dispensmg of reimbursed drugs is recorded:
self-mdication with over-the-counter drugs cannot be measured.
Detailed information legarding the dose of pharmaieutical forms
and the number of umts dispensed aie available. Howevei, the
prescribed dose is not recorded. This should be modehzed thanks
to pharmacoepidemiological methods. These approaches use most
frequently the Defined Daily Dose defined for each drug by the
World Health Organization [35[. Validation of these pharmacoepi-
demiological methods is an important field of research, for instance
regarding the modelization of drug exposure from rare dispensing
[36] or the impact of cumulative doses [37], Dunng hospital stays,
only data regarding the dispensed costly drugs are available. There-
fore, it is difficult to identify an exposure to cyclophosphamide
for instance. Current generalization of computenzed prescription
in French hospitals may fill this "black hole" of in-hospital drug
exposures. Lastly, let recall that dispensing is not synonymous of
drug intake by patients, but this is a pitfall of most studies on
drugs.
Tous droits rservs l'diteur
Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Lapeyre-Mestre / A. Palmaro /
G. Pugnet / J.-L Montastruc /
L. Sailler
Page 6/7
4. Conclusion
The SNIRR-AM and the EGB offer grt possibihties for medical
research, particularly in Internai Medicine: allowmg the building
of cohorts with prospective recording of data and with unequaled
statistical power, they might contribute to hetier know the epi-
demiology of rare diseases, and may have an impact on guidelines
for theii tieatment. A majoi issue lemains the facilitation of data
linkage with the data of clinical cohorts
Disclosure of interest
The authors dclare that they have no conflicts of interest con-
cerning this article.
Appendix A. Supplementary data
Supplementary data associated with this article (The French
version of this article) can be found, in the online version, at
http://dx.doi.0rg/10.1016/j.revmed.2014.ll.009.
References
[I] Haut Conseil de la Sant Publique Pour une meilleure utilisation des
bases de donnees administratives et mdico-administratives nationales
pour la sant publique et la recherche, 2012 http //www.hcsp.fr/explore
cgi/avisrapportsdomaine'clefr=25S [Accessed OI July 2014]
[2] Bras P-L. Loth A Rapport sur la gouvernance et l'utilisation des donnes de
sante, 2013 http://www social-sant gouv fr/IMG/pdf/Rapport_donnees_de
sante_2013 pdf |Accessed OI July 2014].
|3] Bases de donnees utilisables Portail pidmiologie-France Available at
hrtps //epidemiologie-france aviesan fr [Accessed Dl July 2014]
[4] Tuppin P, de Roquefeuil L. Weill A, Ricordeau P, Merlire Y French national
health insurance information system and the permanent beneficianes sample
Rev Epidemiol Sante Publique 2010:58 286-90.
(5] De Roquefueil L. Studer A, Neumann A Merlire Y L'<_hantil\on gnraliste
des bnficiaires rerpsentativite, porte et limites Prat Organisation Soins
2009,40 213-23.
|6] Loi n 98-1194 du 23 dcembre 1998 de financement de la scurit sociale pour
1999 Journal Officiel de La Rpublique Franaise 1998.300 19646
[7] Arrt du 19 juillet 2013 relatif la mise en oeuvre du Systme national
d'information mterrgimes de l'assurance maladie. Journal Officiel de La
Rpublique Franaise 2013.0187 13791
|8] Woild Heallh Organization International Classification of Diseases 2014
http (/www who int/classiRcations/icd/en/ [Accessed OI July 2014]
[9] Loi n 2004-810 du 13 aot 2004 relative a l'assurance maladie. Journal Officiel
de La Republique Franaise 2004.190 14598.
[10] Institut des Donnes de Sant Rapport au Paiement, 2013 http //
www institut-des-donnees-de-sante fr/telechargements/RA-lDS-2013bdf pdf
|Accessed OI July 2014]
[ll] Riou C Fresson], Serre J, AvillachP, LeveneutL QuantinC Guide de bonne pra-
tiques permettant d'assurer la confidentialit des donnes de sant recueilles
en milieu hospitalier lors de leur rutilisation des fins de traitement
statistique, 2013 http://emois org/images/EMOIS2014/soumission resp pdt
[Accessed OI July 2014].
[12] Arnaud L, Fagot J-P, Paita M Fagot-Campagna A, Amoura Z. Prevalence and
incidence of systemic lupus erythematosus in France a 2010 nation-wide
population-based study Autoimmun Rev 2014, http'//dxdoi org/10 1016/
j.autrev.2014.08.034
[13] Moulis C, Palmaro A, Montastruc J-L, Godeau B Lapeyre Mestre M, Sailler L.
Epidemiology of incident immune thrombocytopema in France, a nationwide
population-based study Blood 2014,124 3308-15
[14] QuantinC Benzenme E, Hagi M, AuverlotB, AbrahamowiczM CottenetJ tal
Estimation of national colorectal-cancer incidence usmg claims databases J
Cancer Epidemiol 2012,2012.298369
[15] Bannay A. Chaignot C, Blotire P-0 Weill A. Ricordeau P. Alla F, et al Score de
Chanson partir des donnes du Smiram chanes au PMSI faisabilit et valeur
pronostique sur la mortalit un an Rev Epidemiol Sante Publique 2013,61 :S9
|16] Tuppin P. Cuerq A. de Peretti C, Fagot-Campagna A. Danchm N. Juilliere Y et al.
First hospitahzation for heart failure in France in 2009 patient charactenstics
and 30-day follow-up Arch Cardiovasc Dis 2013;106 570-85
[17] ReyG.JouglaE FouilletA, Hmon D Ecological association between a depnva-
Uon index and mortality in France over the penod 1997-2001 variations with
spatial scale degtee of urbamcity, age, gender and cause of death. BMC Public
Health 2009;9'33
[18] Tuppin P, Samson 5 Weill A. Ricordeau P. Allemand H Seasonal influenza
vaccination coverage in France dunng two influenza seasons (2007 and
2008) and dunng a context of pandmie influenza A(H1 N I ) in 2009 Vaccine
2011,294632-7
SANTE2 6564384400503

Pays : France
Priodicit : Mensuel
[19] Fagot J-P Boutrelle A, Ricordeau P, Weill A, Allemand H HPV vaccination in
France uptake costs and issues for the National Health Insurance Vaccine
2011,29 3610-6
[20] Moulis G Lapeyre-Mestre M, Adoue D, Sailler L Lexposition aux vaccins
recommandes est faible chez les patients non splnectomises atteints de
thrombopnie immunologique primaire et exposs au ntuximab Resultats de
l'tude FA1TH (French Adult Immune Thrombocytopema a pharmacoepidemi-
ological study) Rev Med Interne 2014,35 A78-9
[21] Haute Autorite de sant Purpura thrombopemque immunologique de
l'enfant et de l'adulte Protocole national de diagnostic et de soins, 2009
http //www has-sante fr/portail/upload/docs/application/pdf/2009-12/ald 2
pnds_pti imune cnft adulte.web pdf [Accessed OI July 2014]
[22] Fautrel B Joubert J-M, Cukierman G, Laurendeau C Gourmelen J Fagnani
F Rheumatoid atlantis (RA) comorbidities and biological agents uptake in
France analysis of a national claims database Ann Rheum Dis 2013,72
337
[23] Bezm J Panente A, Lassalle R, Dureau-Pournm C, Abouelfath A, Robinson P, et al
Use of the recommended drug combmation for secondary prevention after a
first occurrence of acute coronary syndrome in France Eur J Clin Pharmacol
2014,70 429-36
[24] Dupouy J Fournier J-P Jouanjus , Palmaro A Poutram J-C Oustric S
et al Baclofen for alcohol dependence in France incidence of treated
patients and prescription patterns a cohort study Eur Neuropsychopharmacol
2014,24 192-9
[25] Gallmi A, Andneu S, Donohue JM, Oumouhou N, Lapeyre-Mestre M, Gardette
V Trends in use of antipsychotics in elderly patients with dementia
impact of national safety warnings Eur Neuropsychopharmacol 2014,24
95-104
[26] Panente A, Pmet M, Monde Y, Merlire Y Moore N, Fourner-Rglat A Fac-
tors associated with persistence of cholmesterase mhibitor treatments in the
elderly Pharmacoepidemiol Drug Saf 2010,19 680-6
[27] Neumann A, Maura G Weill A, Ricordeau P Alla F, Allemand H Comparative
effectiveness of rosuvastatin versus simvastatm in pnmary prevention among
Tous droits rservs l'diteur
Date : JUIN 15
Page de l'article : p.51-57
Journaliste : G. Moulis / M.
Lapeyre-Mestre / A. Palmaro /
G. Pugnet / J.-L Montastruc /
L. Sailler
Page 7/7
new users a cohort study in the French national health insurance database
Pharmacoepidemiol Drug Saf 2014,23 240-50
[28] Weill A, Paita M, Tuppin P, Fagot J-P, Neumann A, Simon D, et al Benfluorex
and valvular heart disease a cohort study of a million people with diabetes
melhtus Pharmacoepidemiol Drug Saf 2010 19 1256-62
[29] Neumann A Weill A, Ricordeau P, Fagot JP Alla F, Allemand H Pioglitazone and
risk of bladder cancer among diabete patients in France a population-based
cohort study Diabetologia2012,55 1953-62
[30] Blin P, Lassalle R, Dureau-Pournm C Ambrosmo B Bernard MA Abouelfath A,
et al Insuhn glargme and risk of cancer a cohort study in the French National
Healthcare Insurance Database Diabetologia 2012,55 644-53
[31] Blotiere P-0 Weill A Ricordeau P, Alla F, Allemand H Perforations and
haemorrhages after colonoscopy in 2010 a study based on comprehen-
sive French health insurance data (SNI1RAM) Clin Res Hepatol Gastroenterol
2014,38 112-7
[32] Dixon WC Abrahamowicz M, Beauchamp M-E, Ray DW, Bernatsky S, Suissa
S, et al Immdiate and delayed impact of oral glucocorticoid therapy on risk
of senous infection in older patients with rheumatoid arthntis a nested case-
control analyse Ann Rheum Dis 2012,71 1128-33
[33] Dixon WC Kezouh A Bernatsky S, Suissa S The influence of systemic glu-
cocorticoid therapy upon the risk of non-senous infection in older patients
with rheumatoid arthntis a nested case-control study Ann Rheum Dis
2011 70 956-60
[34] Moulis G, Sailler L, Adoue D, Lapeyre-Mestre M Pharmacoepidemiology of
immune thrombocytopema protocols of FAITH and CARMEN studies Therapie
2014,69 437-48
[35] World Health Organization ATC/DDD Index, 2014 [Accessed OI July 2014]
http //www whocc no/atc dddjndex/
[36] Pottegrd A, Hallas J Assigning exposure duration to single prescriptions by
use of the waiting time distribution Pharmacoepidemiol Drug Saf 2013,22
803-9
[37) Sylvestre M-P, Abrahamowicz M Flexible modehng of the cumulative
effects of time-dependent exposures on the hazard Stat Med 2009,28
3437-53
SANTE2 6564384400503