White Paper HDI_big data and prevention_EN_Nov2016

1
HEALTHCARE
DATA INSTITUTE
BIG DATA
AND PREVENTION
FROM PREDICTION TO
D E M O N S T R AT I O N
NOVEMBER 2016
INTERNATIONAL THINK TANK
DEDICATED TO BIG DATA IN HEALTHCARE

2
Acknowledgements to the authors of this White Paper:
• Isabelle Barbier-Feraud, Pagamon
• Jeanne Bossi Malafosse, Lawyer
• Patrice Bouexel, TeraData
• Catherine Commaille-Chapus, Open Health Company
• Anne Gimalac, Aston Life Sciences
• Guillaume Jeannerod, Epiconcept
• Magali Léo, CISS
• Bruno Leroy, Sanofi
• Bernard Nordlinger, AP-HP, Académie de Médecine
• Michel Paoli, InterMutuelles Assistance
• Pablo Prados, Sanofi
• Jean-Yves Robin, Open Health Company
BIGDATA
ANDPREVENTION
FROM PREDICTION TO
DEMONSTRATION

3
INTRODUCTION
T
aking action to prevent epidemics, identifying risk factors
as soon as possible in order to avoid diseases or slow their
development, predicting predisposition to particular infec-
tions from birth, and so on. These are among the goals stated time
and time again by scientists and representatives of the medical
community.
Up to now, the response to these issues has involved the use of tradi-
tional research methods, which have been and are still largely based
on the notion of reproducibility: using a hypothesis that serves as
the basis for the collection of data, an observation makes it possible
to develop a relationship among those data and to deduce new hy-
potheses or new treatments, which will then be reproduced in similar
cases.
For a number of years, this research has been enhanced by other
very diverse sources that also take into consideration the environ-
ment in which the individuals are living: environmental data, so-
cio-professional data, etc.
This situation is now being re-examined by the sudden emergence of
new data management techniques, which must adapt to the massive,
exponential production of data characterised by the phenomenon
of Big Data (we will come back to a definition of this term a bit later).
Can this massive production of data have an effect on disease pre-
vention? Does Big Data have a role to play, among other technolo-
gies, in improving the health of the population and thus providing

4
politicians with the right information at the right time so as to enable
the definition of a more intelligent public health policy?
Thedevelopmentof digitaltechnologiesoverrecentyearshasalsore-
sulted in a veritable 'datification' of our society. Data are everywhere,
and constitute a valuable raw material in the development of new
knowledge and a major global factor for growth for a number of
countries.
This development is also characterised in the health sector by the
increasing computerisation of the professional sector, in particular
activities associated with care and prevention, life sciences research,
and health systems management, and by the growing involvement
of patients.
The phenomena of the mobility and development of connected
medical objects and devices are also contributing to the exponential
growth of the volume of data produced, and the phenomenon of Big
Data merely reflects the specificities involved in the processing of
large volumes of data, with the associated requirements for speed of
processing, uniformity of the data and specific value creation.
Health data therefore represent a unique tool and a potential to
create value, with the realisation of that potential depending on the
ability of the various nations to organise the development of an eco-
system that facilitates the use of the data while guaranteeing compli-
ance with privacy requirements and the confidentiality of personal
data.
Indeed, governments are currently being asked to address major is-
sues in which the processing of health data can play and is already
playing an essential role: questions of public health, questions of
quality of care, questions of transparency and health democracy,
questions of efficiencies within health systems in a general context
of increasing health expenditure, and questions of innovation and
growth in areas as varied and significant as personal medicine or
precision medicine and information technologies.
More than two years ago, the Healthcare Data Institute initiated a
process of reflection and discussion on the subject of Big Data and
health, and has put together a working group specially dedicated

5
to the subject of Big Data and prevention. Its goal was to provide
a few examples illustrating the role that Big Data can play in health
prevention, to identify the factors that promote its development and
those that frustrate it, and to propose certain areas for investigation
that could be valuable for decision-making, both private and public.
The work described in this White Paper is the result of the discussions
and activities undertaken by individuals from a range of different pri-
vate and public professional contexts. It is clearly not intended to be
exhaustive on this subject but, rather, to draw attention to domains
that are currently integrating new areas of potential for Big Data and,
at the same time, to highlight certain problems of varying types.

6
DEFINITIONS
I
n order to present this work, there was a need to agree on two
important definitions - the definition of Big Data and the definition
of prevention.
Big Data describes data sets that become so large that they are diffi-
cult to process using only database management tools or tradition-
al information management tools. Big Data also describes all of the
technologies, infrastructures and services that make it possible to
collect, store and analyse data gathered and produced in increasing
quantities, using automated processing operations and artificial in-
telligence technologies1
.
It is standard practice to illustrate this definition using the image of
the '3Vs' that characterise Big Data: the explosion of the volume of
the data, the variety of the structured and unstructured data pro-
duced by multiple sources, and the velocity of the information and
simultaneous processing speed.
A fourth V should be added to these three: the value represented
by these data for the entity or individual. Data have become the raw
material of the digital universe.
It is easy to see the confusion that often arises between the term Big
Data and the usage of data analysis methods such as Data Mining. As
a result of this confusion, the term Big Data is used very extensively
to describe all new data analysis methods.
The working group however deliberately made the decision to use
this broad definition of the term Big Data in order to adapt the scope
of its analysis to the actual realities of the subject: the question we
need to address today is more about knowing how to create intelli-
gent databases in order to manage Big Data.
So, in this White Paper, the term Big Data will be used flexibly to
group together all of the techniques and methods currently used to
1. Report from the Institut Montaigne on Big Data and Connected Objects, 2015.

7
analyse the increasingly large volumes of data produced by a variety
of sources.
With regard to prevention, which is also a very vast term, we have
used the definition provided by the World Health Organisation
(WHO). According to that institution, prevention is 'all of the meas-
ures intended to avoid or reduce the number or severity of diseases
or accidents'.
Three types of prevention are then identified:
• Primary prevention, which covers 'all of the actions intended to re-
duce the incidence of a disease, and therefore reduce the frequency
of new cases'.This involves the use of individual prevention measures
(personal hygiene, food, etc.) and/or collective prevention measures
(distribution of potable water, vaccination, etc.).
• Secondary prevention, which covers 'all actions intended to reduce
the prevalence of a disease and therefore reduce its duration'. This
includes screening and the treatment of new outbreaks.
• Tertiary prevention, which concerns 'all actions intended to re-
duce the prevalence of chronic disabilities or relapses in the pop-
ulation and therefore reduce the functional disabilities caused by
the disease'. The aim of these strategies is to promote social and
professional reintegration following the disease. This definition
extends the notion of prevention to cover rehabilitation treatment.

8
TABLE OF CONTENTS
The subjects addressed in this White Paper by means of a series
of articles are organised around several horizontal and transverse
themes. Each article can be read independently from the others.
I. An initial theme addresses the role of Big Data and personalised
prevention on the basis of the issue of genomics.
II. The second theme discusses universal population prevention and
illustrates the possibilities for the management of real-life data to
benefit public health and the prevention of health risks.
III. The third theme selected covers the essential subject of the de-
velopment of the technological landscape as a key success factor for
'making the data speak'.
IV. The fourth theme addresses the question of the definition of a
new paradigm for the stakeholders involved.
The choice of these four subjects has made it possible for the mem-
bers of the working group to identify the common, cross-cutting
problems in relation to each theme that appeared to warrant analysis
and could serve as the basis for recommendations.
This is covered in Article V, in relation to the important issue of the
anonymous or pseudonymised nature of the data used and the ap-
propriate degree of aggregation to be applied, and Article VI on
ethical and legal aspects, which are covered in this document es-
sentially through the theme of patients and making data accessible.

9
ARTICLE 1 Big Data and genomics:
what future do they have in the treat-
ment of cancer? p. 10
1. Context and technology p. 10
2. Scientific research
and the medical treatment of cancer
today p. 11
3. How can mass genomic data analysis
contribute in this mechanism? What
can we expect from Big Data? p. 13
4. What are the obstacles to overcome
in achieving 6P medicine based in
particular on Big Data
and molecular diagnosis? p. 15
> a. Technological challenges p. 15
> b. Organisational challenges p. 16
> c. Political and
economic challenges p. 16
ARTICLE II Data and population health
prevention p. 18
1. Opportunities for the use of
Big Data in pharmacovigilance p. 20
2. Big Data in the prevention
and monitoring of epidemics p. 21
3. Cross-referencing of clinical and
socio-demographic data: moving
towards an improved predictive factor
for treatment compliance? p. 24
4. Big Data and the personalisation
of prevention p. 26
5. Conclusion: methodological
and ethical factors p. 28
ARTICLE III Development of the tech-
nological landscape as a key success
factor in 'making the data speak' p. 30
1. Introduction p. 30
> a. Variety p. 33
> b. Volume p. 33
> c. Velocity p. 35
> d. The 'Volume/Velocity/Value' ratio p. 36
> e. Variability and veracity p. 37
> f. Visualisation p. 39
> g. Current difficulties p. 40
> h. Conclusion p. 40
> i. A unifying infrastructure – essential
for Big Data p. 40
> j. Glossary p. 47
2. 'Knowledge by design' or semantics
as the key to value p. 49
ARTICLE IV The definition of a new
paradigm for stakeholders p. 52
1. Big Data: a new paradigm
for the pharmaceutical industry p. 52
> a. Patient monitoring and individual
secondary prevention –
chronic diseases and digital health p. 54
> b. Examples of initiatives for integrated
health solutions using real-time
data analysis p. 56
2. The insurance sector
and Big Data: predictive and therefore
personalised health? p. 57
> a. Background p. 57
> b. Impacts and perspectives p. 59
ARTICLE V Anonymised data,
pseudonymised data:
what degree of aggregation? p. 67
ARTICLE VI Releasing data:
patients, users at the heart
of the 'disruption' p. 71

10
ARTICLE 1
BIG DATA AND GENOMICS: WHAT FUTURE DO
THEY HAVE IN THE TREATMENT OF CANCER?
1. CONTEXT & TECHNOLOGY
It has been said that we are entering the era of '6P Medicine' (Person-
alised, Precision, Participative, Preventive, Predictive and Patient-ori-
ented) based on a genomic and molecular approach to disease that
provides an opportunity to:
• better understand pathological mechanisms;
• identify new therapeutic targets;
• identify risk factors;
• support diagnosis and medical decision-making;
• personalise treatment.
What is this?
For a given condition, the sequencing of all or part of a patient's
genome makes it possible to identify the orthographic differences
(referred to as variants) compared to the genome of individuals who
are not suffering from the condition in question. Over recent years,
considerable technological advances in the area of high-throughput
sequencing (genome decoding method) have made it possible to
reduce the associated times and costs and to facilitate the storage
and analysis of the mass genomic data. This could suggest that the
sequencing of the entire genome (and not just certain target sec-
tions) for individuals and patients could soon be a matter of routine,
paving the way for applications in the domains of both research and
treatment.
But where actually are we at the moment? Can we really apply Big
Data to the realm of genomics? Will we reach that point soon or will
it take us longer? What will be the outcome and which obstacles will
we need to overcome to get there?

11
In the remainder of this section, we have chosen to illustrate the
example of cancer.
2. SCIENTIFIC RESEARCH AND THE
TREATMENT OF CANCER TODAY
Considerable progress is currently being achieved in the understand-
ing and treatment of cancer, based both on a better understanding
of the impact of environmental factors, on clinical observation, on
clinical trials of new pharmaceutical treatments, and on progress in
immunotherapy (in particular for melanoma) and biopsies on fluids
(and not merely tissues).
We are also seeing the development of the genome analysis of tu-
mours themselves and of patient genomes. By way of example, at the
French National Cancer Institute (INCA), 70,000 new patients each
year are having their genome subjected to the targeted sequencing
of certain regions, followed by routine analysis in order to identify
any molecular anomalies and to personalise their treatment in or-
der to increase efficacy and reduce side effects. It is expected that
whole-genome sequencing will be routine by 2020, and this should
therefore facilitate a systemic understanding of the disease.
Furthermore, we are seeing a change in the way new molecules are
being developed from academic and industrial research, and in how
the efficacy of those molecules is demonstrated during regulated
clinical trials. Patients recruited for Phases I, II and III clinical trials
can now be selected on the basis of their genome profile. Previous
Phase III studies are, in particular, being reanalysed in the light of
the discovery of new biomarkers (certain specific genetic markers
expressed by the tumour) in order to evaluate the benefit of targeted
therapies for tumours in patients with a genome profile associated
with a better response to the treatment, in other words those who
make it possible to predict how a tumour will respond to a given
treatment. The idea is to have access to data that enable the admin-
istration to a patient of the treatment that is best suited to his or her
disease, in order to increase their chance of recovery. It has now,
in fact, been clearly established that not all patients suffering from

12
cancer react to a treatment in the same way. Response depends on
both the specific characteristics of the patient and the specific char-
acteristics of the tumour. We are therefore researching biomarkers
correlated to a better response to a treatment and biomarkers that
make it possible to estimate the risk of relapse.
Prognostic biomarkers are biomarkers that allow us to better pre-
dict the lifetime or lifetime without relapse for the patient, as another
means of improving treatment.
In the case of colorectal cancer, we have been able to establish that
only patients whose tumours have a normal (non-mutated) version
of a gene called RAS are able to benefit from the effect of cetuximab
and panitumumab, two new medicinal products (monoclonal anti-
bodies) for targeted therapy. In this case, predictive biomarkers are
biomarkers that make it possible for us to better predict the efficacy
of a treatment.
But here, we still cannot really consider this to be a Big Data ap-
proach to genomics, in particular in terms of mass data or diversity
of sources. Although we now know how to cross-reference the ge-
netic information characterising the tumour and the response to a
therapy for several hundreds or thousands of individuals, the selec-
tion of treatment based on genome sequencing has still not become
systematically routine in medical practice.
At present, the application of Big Data in genomics only occupies a
very small space among the methods for research and treatment of
disease, and cancer in particular.
There are however certain isolated initiatives that are moving in this
direction, including the following examples (non-exhaustive list): in
England (The 100,000 genomes project); in Germany (systematic
genotyping of tumours at the DKFZ - German Cancer Research Cen-
tre in Heidelberg, a model to be applied); in the United States (Can-
cerLinQ, Cancer Moonshot); in France (France Médecine Génomique
2025, INSERM/Quest Diagnostics 'BRCA Share' to improve diagno-
sis of predisposition to breast and ovarian cancer); in Switzerland
(Personalised Health); in Europe (SPECTAcolor, Million European
Genomes Alliance, Sophia Genetics*), etc.

13
3. HOW CAN MASS GENOMIC
DATA ANALYSIS CONTRIBUTE
IN THIS MECHANISM?
WHAT CAN WE EXPECT FROM BIG DATA?
In epidemiology, the analysis of mass data, cross-referencing the
genetic/genomic data about patients and their tumours with ex-
ternal environmental data (geography, diet, atmospheric pollution,
UV radiation, etc.), connected health data, etc., would considerably
improve the effectiveness of current treatments, making it possible
to better identify the risk factors that influence cancer survival and,
therefore, contribute to their prevention.
Any progress in this area will depend on the possibility of cross-refer-
encing clinical and genetic data on tumours and patients, and there-
fore on the creation of biobanks and prospective and retrospective
databases, and access to those tools that is unrestricted but strictly
regulated in terms of protection of medical data. We will have gen-
*It is interesting to observe a new trend for the application of Big Data,
at the boundary between biotechnology, genomics, digital health and
artificial intelligence. Using sophisticated algorithms, 170 hospitals in
28 European countries are currently using the services associated with
analysis and interpretation of the genome profiles of their cancer pa-
tients (on the basis of pooled data) in order to support decision-making
by their doctors in relation to the degree of pathogenicity of their ge-
netic mutations. The goal is not for the algorithm to replace diagnosis
by a physician but, rather, to accelerate the diagnosis of 'simple' cases
in order to free up the doctor's time for the diagnosis of more com-
plex cases that involve the accumulation of a range of different genetic
mutations.

14
uinely entered the era of Big Data in genomics when we are able
to accumulate genomic and diagnostic data on large numbers of
patients, clinical data before/during/after treatment, etc., that make
it possible to define potential correlations between genotypes and
phenotypes and thus to predict the influence of environmental fac-
tors and susceptibility to diseases or therapeutic incidents. These
Prof. Pierre-Laurent Puig, speaking before the Working Group, physi-
cian in the Biochemistry Department of Hôpital Européen Georges
Pompidou (Paris) and Director of the INSERM UMR-S775 Research
Unit at the Université de Médecine Paris Descartes. Research subject:
Colorectal cancers.
'Today, the generation of genomic data is easy and not costly. In terms
of application, we can no longer dissociate scientific research from med-
ical treatment, because we are analysing the associated raw sequencing
data in order to characterise a specific medical condition and adapt the
treatment of patients suffering from that condition. However, in France,
we are facing a regulatory obstacle (not technological or financial) as-
sociated with the protection of health data, which does not authorise
the cross-referencing and correlation of genomic data with clinical data
from medical files, and does not authorise the use of data for purposes
other than those for which they have been generated and collected. The
databases are created retrospectively. The prospective approach, bring-
ing together data collected without a central purpose, will be possible
when we are able to put in place electronic medical records that are cen-
tralised from the outset and created for the purposes of management
and research without any pre-established purpose.'

15
prospective and retrospective databases will be dedicated to re-
search, diagnosis and treatment, and could make it possible to ad-
dress the questions raised, eliminating the requirement associated
with the initial purpose for which the data are collected.
4. WHAT ARE THE OBSTACLES TO OVERCOME IN
ACHIEVING 6P MEDICINE BASED IN PARTICULAR
ON BIG DATA AND MOLECULAR DIAGNOSIS?
Please find listed below what we believe to be the primary
technological, organisational, political and economic challenges.
A. TECHNOLOGICAL CHALLENGES
The software and algorithm market requires:
1) opening up to digitised/digital data (essential for traceability,
even if this creates new risks);
2) professional knowledge;
3) standardisation, repeatability.
It is therefore necessary to:
• Develop standardised electronic patient files incorporating genom-
ic data and traditional clinical data.
• Create the ecosystem that makes it possible to bring together the
skills of engineers, IT technicians, molecular geneticists, geneticists,
etc.
• Develop algorithms, applications and processes to ensure confi-
dentiality and security of data so as to enable medical practitioners
to provide a clear opinion, diagnosis and treatment regimen.
• Have the algorithms validated by all of these experts so as to guar-
antee that they are robust, easy to use and applicable (IBM Watson is
only applicable for American medical practice!).
But also to:
• Put in place a multidisciplinary network to address the exponential

16
requirements for high-throughput sequencing, collection, storage,
large-scale analysis, and decoding of the databases to ensure that
they can be understood by the practitioner.
B. ORGANISATIONAL CHALLENGES
• Guarantee the confidentiality of the genetic data and ensure com-
pliance with a regulatory framework so as to avoid discrimination.
• Prepare doctors for the coming genome revolution, train them in
changing practices, prepare them for the decompartmentalisation
of medical specialisms (as the disease will no longer be viewed in
terms of its location within the body but, rather, at a molecular and
cellular level).
• Develop new tools with the practitioners themselves in order
to make them ambassadors contributing to the development of
medical practice and the design of the medicine of the future.
• Implement work in synergy with other professions (bio-IT, statistics,
etc.) to determine the most appropriate protocol.
• Develop the doctor/patient relationship to support patients ex-
cluded from sub-groups, for which a medicinal product has been tar-
geted, to support patients with significant genetic predisposition, to
educate patients in new preventive behaviours, and to ensure a bet-
ter understanding by the patient, thus enabling informed consent.
• Reorganise scientific and technological research on the basis of the
development of medical and clinical practice, consider new public/
private arrangements, etc.
C. POLITICAL AND ECONOMIC CHALLENGES
• Develop the health system, determine the conditions and levels of
treatment for health insurance.
• Create an industry sector around genome and digital health, to
mobilise the sequencing, data storage and scientific instrument
industry, and to design a new model for the pharmaceutical industry.

17
• Bring together stakeholders such as the pharmaceutical industry,
with teaching hospital institutions, as early as possible in the value
chain, in order to provide access to genomes from cohorts of pa-
tients and identify the biomarkers associated with various conditions.

18
ARTICLE II
DATA AND POPULATION
HEALTH PREVENTION
In health as in other fields, technological progress has caused an ex-
plosion in the quantity of information collected every moment. The
use and analysis of these growing volumes of data available from a
variety of different sources that collect them for various reasons are
a source of considerable hope and progress in terms of scientific
advances, and create new opportunities for population health pre-
vention and the management of health risks.
Population health prevention is one of the roles of public health, de-
fined as the 'science and art of promoting health, preventing dis-
ease, and prolonging life through the organised efforts of society2
'.
In this interpretation, prevention covers all measures intended to
avoid the occurrence of diseases and accidents or to reduce their
number, severity and consequences.
Big Data are based on the ability to collect, aggregate and process
data originating from a wide range of heterogeneous sources. The
variety of data collected (structured, unstructured, etc.) and sources
(public/private databases, medical and administrative data, health
data/environmental data, etc.), their volume, and the velocity with
which they are gathered and processed are at the heart of approach-
es based on Big Data. The spread of the use of connected objects
and the development of geolocation practices are contributing to
the significant growth in the volume of available data. The analysis of
these large-scale data, which is made possible both by the increas-
ing digitisation of activities within society and by the much-enhanced
capacities for storage and processing, is revolutionising traditional
approaches to prevention.
The analysis of data has always been at the heart of our understand-
ing of health phenomena. The knowledge gained by the public
2. D Nutbeam, WHO, 1998.

19
authorities, which is necessary for action to be taken, is historically
based on activities involving epidemiological observation, which are
intended to observe and measure the health-related events affecting
a given population, to explain them, and to determine the impact of
the measures taken to address them.
What is new today using the techniques associated with Big Data is
this new ability to aggregate and process massive volumes of data
originating from different sources, coupled with the fact that there is
now a considerable bank of digital data available, both in the health
sphere (treatment data, medical and administrative data, data pro-
duced by patients themselves using connected objects, etc.) and
more widely about socioeconomic, cultural and environmental con-
ditions, which are the key determinants for population health, such
that there is no longer necessarily any need to produce these data
specifically. The European Regulation on the protection of personal
data, adopted on 27 April 2016, thus enshrines, for the first time, the
concept of 'compatible purpose', authorising the reuse of data for a
purpose other than the purpose for which they were initially collect-
ed, provided that the intended purpose is compatible with the initial
purpose.
While up to this point, health-related studies have required the col-
lection of ad hoc data, on the basis of criteria used to populate data-
bases, often over very long periods, data produced during treatment
activities or for the purpose of reimbursement of costs, or even by
patients themselves using connected objects or social networks, are
now available and constitute an almost inexhaustible source for the
identification of disease risk factors, health security or epidemiology.
The processing and analysis of these still largely under-used big data
provide limitless opportunities for knowledge generation. These
analyses can contribute to the prevention of diseases and epidem-
ics and to surveillance and medical supervision, making it possible
to better understand the socioeconomic and environmental deter-
minants for population health, to detect unusual health events that
could represent public health alerts, and where necessary to es-
tablish links with exposure-related factors. They can make it possi-
ble to target prevention efforts towards the population groups for
which these measures are most effective, and can contribute to the

20
monitoring and evaluation of public health actions.
The following sections illustrate the variety of possibilities provided
by Big Data in terms of population health prevention, supervision
and health security.
1. OPPORTUNITIES FOR THE USE OF BIG DATA
IN PHARMACOVIGILANCE
In a country that is classified among the biggest consumers of me-
dicinal products in Europe, the question of the adverse effects asso-
ciated with their consumption and the monitoring of correct use is
increasingly becoming a major public health issue. The recent health
crises (Mediator, 3rd
and 4th
generation pills, Depakine, etc.), the im-
pact of which is clearly considerable in terms of public health and
in financial terms, illustrate the difficulties that the health authorities
and the national community must address in relation to the misuse of
drugs that is more extensive than in the past, combined with a lack of
precise information about the conditions for use of drugs3
. Allowing
misuse to continue or discovering the serious effects of this phenom-
enon ten years after the fact is both dramatic in terms of population
health and extremely costly for the community. In European terms,
the cost of adverse effects linked to drugs is estimated as 197,000
deaths per year in the EU. The financial impact is valued at 79 billion
euros per year4
.
The monitoring of the safety of drugs is based essentially on the
evaluation actions that can be undertaken following market introduc-
tion and dissemination within the general population, in population
groups that are broader and more diverse in terms of their charac-
teristics than those included in clinical trials, and for which the use
of the drug (treatment period, posology, observance, etc.) is itself
3. B. Bégaud, D. Costagliola, Report on the monitoring and promotion of the correct use
of drugs in France [in French], pharmaco-monitoring project entrusted by the Minister of
Social Affairs and Health, Ms Marisol Touraine, 26 February 2013.
4. H. Pontes, M. Clément, V. Rollason, Safety Signal Detection: The Relevance of
Literature Review, Spinger International Publishing Switzerland 2014.

21
much more variable5
. The analysis of the data from patient cohorts or
medical and financial databases over the long term can make it pos-
sible to detect signs and establish links between the occurrence of a
health event and exposure to a given treatment, and can trigger an
alert as to the possible adverse effects of drugs or off-label use not in
accordance with the indication for which a given health product has
been awarded a marketing authorisation.
The speed of access to the relevant knowledge is absolutely key.
In the case that serious adverse events are detected, the periods in
years that can elapse between the suspicion of a phenomenon and
the establishment of a sufficient level of proof can be highly dam-
aging. Big Data-type search techniques can be applied to support
these analyses and can enable the early detection of warning signs
that are decisive in public health terms.
2. BIG DATA IN THE PREVENTION
AND MONITORING OF EPIDEMICS
The availability of large-scale data from different sources, in real-time
or near-real-time, means that information can be obtained about
the health status of a population in a given geographical zone. The
cross-referencing and analysis of these data, combined with mathe-
matical modelling techniques, can make it possible to identify trend
shifts that herald an increase in the incidence of diseases or behav-
iours and suggest that a probable change in the health status of that
population can be expected.
Thus, alongside traditional information gathering and alert mecha-
nisms for infectious diseases and epidemics (such as the Sentinelles
network in France, made up of doctors throughout the country who
report the cases observed each week for a certain number of trans-
missible diseases), epidemiological models are being developed to
monitor the space-time propagation of health-related phenomena
such as epidemics through the use of large-scale data.
5. J.L. Faillie, F. Montastruc, J.L. Montastruc, A. Pariente, The contribution of
pharmacoepidemiology to pharmacovigilance, 2016.

22
A little over ten or so years ago, the use of Big Data was initiated
within the new public health agency to bring together the InVS, the
INPES and the EPRUS, which was entrusted with the task of protect-
ing the health of the French population. The associated processes
were based above all on the use of data from defined, known data
sources.
Of the 3Vs frequently associated with Big Data, Variety is most
important.
By using stable, validated data sources such as PMSI, SNIIRAM,
CepiDC or data from cancer registers, scientifically validated in-
dicators that have been shown to be robust over time have been
developed.
SurSaUD (the system for the health surveillance of emergency de-
partments and deaths), the monitoring system put in place following
the 2003 heatwave and based on data from A&E departments, SOS
Médecins, mortality data and death certificates, is just one example.
These indicators make it possible to produce the BQA (Daily Alert
Bulletin) intended for the health authorities, including the Minister
of Health.
A system of this kind is highly valuable in detecting unexpected
health-related events, estimating the impact of an environmental
or societal event, monitoring medical conditions where no events
have been detected, or enabling the early detection of a predefined
health-related event, such as a seasonal epidemic, and in measuring
the impact and consequences of that event.
While the various systems put in place are effective in detecting an
event or confirming a trend, they have limited or nil capacity to ex-
plain the associated causality within the deadlines imposed by the
need to respond to these situations and by political pressure.
For information purposes, in the Île-de-France Region in March 2014,
an increase in asthma was observed and attributed initially to the
peak in pollution that was rife at that time. Subsequently, it became
clear that a larger quantity of pollen than usual was circulating in the
air at that time and that the pollution was merely a third factor.
The example of Google Flu Trend, which was intended to monitor

23
the development of the flu around the world on the basis of an anal-
ysis of certain key words within the search engine, has been repeat-
edly highlighted. A comparative analysis of detection methods with
other methods for monitoring the propagation of the flu syndrome
has shown the limitations of this model in respect of determining the
extent of the phenomenon. Google was forced to reintegrate data
originating from the American Centres for Disease Control (CDC)
and the service has since been closed.
The use of data from mobile telephones is a very interesting ex-
ample, as these data can in fact make it possible to provide a very
precise description of the effects that can be generated by public
gatherings and movements on the propagation of an epidemic such
as cholera or malaria. An understanding of these phenomena, which
enable the identification of 'transmission points' for an epidemic,
has huge potential in terms of contributing to the eradication of
infectious diseases6
.
Other initiatives, such as an analysis of the browsing frequency for
certain Wikipedia pages (such as the page on bronchiolitis) or an
increase in keywords on Twitter, enable the detection of weak signs.
While it is important, as these examples show, to remain cautious
in terms of the validation of these analyses, which require observa-
tions gathered on the basis of long case series over time and over
distance, and while these models have not yet shown their full po-
tential, the fact remains that they offer excellent prospects in terms of
the prevention and control of pathologies, and especially infectious
conditions.
The power of the algorithms and an increase in sources of data over
time will provide a guarantee that every health-related event will be
detectable.
However, it is and will always be essential to allow for the time and
the medical expertise provided by the men and women in the field
to triage the vast quantity of events detected and understand and
6. Flavio Finger, Tina Genolet, Lorenzo Mari, Guillaume Constantin de Magny, Noël
Magloire Manga, Andrea Rinaldo and Enrico Bertuzzoa, Mobile phone Data highlights
the role of mass gatherings in the spreading of cholera outbreaks.

24
ensure a reliable matching between a health-related event and its
cause. This is not necessarily compatible with the information society
that is developing at the same speed as these technologies.
3. CROSS-REFERENCING OF CLINICAL AND
SOCIO-DEMOGRAPHIC DATA:
MOVING TOWARDS AN IMPROVED PREDICTIVE
FACTOR FOR TREATMENT COMPLIANCE?
Socio-geographical data, traditionally not used in the health sphere,
could be integrated into the design of more personalised pub-
lic health intervention programmes, by means of modelling using
advanced machine-learning techniques.
These techniques make it possible to develop models for predict-
ing treatment compliance and the onset of complications such as
acute cardiovascular events or episodes of severe hypoglycaemia
in patients suffering from chronic diseases such as diabetes. The
public health intervention programmes could then be optimised by
giving priority to the use of certain resources for the population cat-
egories most exposed to these complications, based on their risk
of non-compliance with treatment assessed on the basis of certain
geographical characteristics or characteristics of socioeconomic
behaviour. Models similar to those used in mass consumption to
predict purchasing behaviour are currently being tested with this
in mind. Most systems of recommendations make suggestions of
personalised choices based on the experience drawn from previ-
ous purchasing behaviours. These approaches have not yet been
used extensively in public health and represent a possible source of
improvements in the quality of care.
This approach firstly requires that reimbursement data be linked
anonymously to demographic medical data and data on the course
of the disease, in order to identify the predictive factors for treatment
compliance and for complications associated with the diseases con-
cerned. The socioeconomic and geographical data are then added
to the analysis to compare or improve that prediction and thus guide

25
the intensification or simplification of the intervention programmes.
Numerous socioeconomic and geographical data are now available
freely in certain countries such as the United States (census data,
consumer expenditure figures, national salary study, food environ-
ment, housing survey, etc.). One critical point, nonetheless, is the size
of the region concerned, given the need to ensure the anonymity of
the medical data. Data on patient behaviour can also be obtained
using digital applications, enabling, for example, the monitoring of
social and physical activity using a smartphone, collected directly or
using simplified questionnaires administered regularly.
The types of data that can be analysed, the associated opportunities
and the possible impact on patient management are summarised in
the graphic above.
DATA ANALYSIS RESULTS BENEFITS
Prediction of
behaviour /
commitment
Stratification for
multi-channel
intervention on
compliance
Prediction of
events /
complications
Longitudinal
assessment of
symptoms
Clinical decision
Identification of
populations at risk of
complications
Public health
programmes
Adaptation of
treatments in
real time
Integrated solutions
(drugs, medical
equipment, support
for treatment)
Aggregation
of new data
New
modelling
techniques
TRADITIONAL
• clinical studies
• observational studies
• registers
• pragmatic studies
• databases of treatment
or insurance records
NON-TRADITIONAL
• socio-geographical
• behaviour
• genomic
• sensors
• social network
• support programmes
• targeted questions on
smartphone
• games on smartphone
+

26
4. BIG DATA AND THE PERSONALISATION
OF PREVENTION
The field of prevention covers both collective actions intended to
protect the health of persons through the development of a physical
and social environment that is favourable for health (health monitor-
ing, vaccination, screening for certain diseases, preservation of air
and water quality, etc.) and the promotion of individual behaviours
that are favourable for health (prevention of alcoholism and smok-
ing, promotion of physical activity and nutritional advice ('eat well,
be active'), etc.).
Prevention activities are therefore aimed at the entire population
and exist in all places in the lives of individuals: at home, at work, at
school, in the city, etc.
Historically, prevention programmes have been implemented by
means of large-scale actions most often referred to as 'campaigns':
• vaccination campaign;
• screening campaign;
• campaign to raise awareness about at-risk behaviours, etc.
These campaigns, conducted by the organisations responsible for
population health, are very non-specific, in the sense that they are
intended to reach the largest possible number of people, often on
the basis of cost/effectiveness analyses. The individual benefit is then
a consequence of the benefit for the population:
• Providing information for the entire population by means of a
radio or televised message in order to launch a vaccination
campaign against the flu while focusing on the elderly population.
• Proposing a mammogram every two years in all cases for all
women aged 50 to 74 years.
In the two examples cited above, we can see that the campaigns are
aimed at a very broad audience in order to increase the impact on the
population. There is little targeting, and the persons concerned are
supposed to recognise themselves in the mechanism put in place.
For screening campaigns, in the case of a positive test, medical

27
follow-up and potential treatment will be customised as part of a
personalised care path downstream from the screening activities.
Today, the availability of the data and the intelligence that can be ob-
tained by cross-referencing those data make it possible to aim for the
largest possible number of individuals while targeting the messages,
redirecting people who are at risk or excluding those not affected.
By using the same techniques as those used on the internet for
advertising, it is possible to get the right message to the right per-
son, taking into account various items of 'highly personal' informa-
tion: medical history, drug consumption, genetic information or
behaviour.
The more these data are available, the more specifically the cam-
paign will be targeted and the more precise and tailored the 'pre-
vention proposal' will be.
These possibilities offered by Big Data should ensure a significant
improvement in public health programmes.
This promise of greater efficiency and substantial savings has been
clearly understood and integrated by bodies such as suppliers of
connected objects in the area of well-being, the members of GAFAM
(Google, Apple, Facebook, Amazon and Microsoft), the mutual insur-
ance organisations and, more recently, the health insurance sector
(the 'active health' programme).
We can cite the change that is currently taking place in the breast cancer
screening programme. Initially, this was a vertical programme dedicat-
ed to an identified population: women aged 50 to 74 years who were
offered a screening test every two years. This programme will be opti-
mised as a result of the availability of genetic data from women with
hereditary predispositions to breast cancer.
The mechanism will be adapted to create a system where systematic,
mass screening becomes 'personalised' mass screening that takes into
account the risk factors specific to each woman, with appropriate per-
sonalised monitoring plans.The French National Cancer Institute (INCA)
is currently working on the introduction of these programmes.

28
Individual coaching can be provided by artificial intelligence systems
on the basis of behaviour. In the context of the mutual insurance
organisations, 'good behaviour' can be rewarded by presentation
of a gift or even a change in the amount of the premium in certain
countries such as the United States.
For the pooling of risks, the founding principle underlying pub-
lic health programmes and health insurance systems in the broad
sense, there will be an increase in the temptation to use models that
will identify 'bad behaviours' by individuals or propose that at-risk
people be excluded from the system.
It is easy to imagine that we could see a move towards elitist, behav-
iourist models promoted by entities managing data for the purposes
of creating profit for society and for themselves.
The question of the roles of the various parties is being asked more
than ever before: governments, institutions responsible for popula-
tion health, experts, companies and bodies that own the data.
For these prevention programmes, the role and responsibilities of
each must be clearly stated and applied so as to avoid any abuses,
which would ultimately conflict with their reason for existing.
5. CONCLUSION: METHODOLOGICAL
AND ETHICAL FACTORS
The cross-referencing of an increasing quantity of health-related and
other data results in the creation of hypotheses relating to health
and the associated environmental and socioeconomic determinants,
and could make it possible to provide access to new knowledge.
These data originate from a wide range of structured and unstruc-
tured databases, using measurement methods that vary depending
on the database and the year, missing data, sources, subject origins
and methods of collection that are extremely varied. The question of
method in this context and in the context of controlling information is
becoming absolutely central for our public system health7
.
7. C. Dimeglio, C. Delpierre, N. Savy, Y. Lang, Big Data and public health: more than ever,

29
Similarly, the availability of the data and the possibilities that this
entails in terms of assessing the effectiveness of public health ac-
tions should lead to new responsibilities for the stakeholders in the
system. The monitoring of data in real time can make it possible to
assess and measure the effectiveness of actions taken and to adjust
them where appropriate. However, it is becoming impossible to ig-
nore the absence of results. This is the case, for example, with the
failure of the campaigns in relation to vaccination against seasonal
flu that are repeated in exactly the same way year after year, while
the results in terms of population coverage remain very much below
the recommendations issued by the WHO. This has consequences in
terms of increased mortality associated with the epidemic each year.
The data, where they are available and able to provide clarification
about the efficiency of the action, detecting signs and monitoring
epidemics and health crises, can no longer remain inert. Their ac-
cess must be open to multiple stakeholders for public health pur-
poses, fostering permeability between the entities responsible for
health policy and the world of data sciences, developing modern
modes of regulation and cooperation between the political sphere
and the stakeholders in the economic world, the drivers of innova-
tion, and supporting the accessibility and use of health data. This is
to take place under conditions that guarantee respect for privacy,
thus strengthening efforts to prevent diseases and improve health
security.
the challenges of knowledge [in French], ADSP Nº 93, December 2015.

30
ARTICLE III
DEVELOPMENT OF THE TECHNOLOGICAL
LANDSCAPE AS A KEY SUCCESS FACTOR IN
'MAKING THE DATA SPEAK'
1. INTRODUCTION
Why is IS architecture providing new
opportunities for 'Mégadonnées8
'?
The technologies specific to Big Data do not, in and of themselves,
provide anything new in terms of the concepts for processing data
themselves. However, they do in terms of the way that they are in-
tegrated in order to, on the one hand, meet the specific needs of
'Volume, Velocity, Variety' and, on the other hand, offer new opportu-
nities in terms of 'Variability, Veracity, Value, Visualisation9
'.
In dealing with Big Data, traditional technologies are limited by their
rigidity in terms of data format ('Variety'), and by the non-scalability
of their tools (Volume), the vast number and growth of available data
sources and the growing number of analysis techniques necessary
('Velocity').
Big Data is ultimately more an evolution than a revolution. Today's
Big Data will be tomorrow's 'Small Data'. The fact remains that the
core is still 'the data'.
If we need to have a definition, the Big Data technological landscape
designates all of the technologies, architectures, infrastructures and
procedures that enable the very rapid capture, processing and anal-
ysis of large quantities of data and changing, heterogeneous content
8. French term recommended by the Official Gazette of 22 August 2014.
9. These requirements are often described as the 'Seven Vs' or the '7 Vs'.

31
in order to extract the pertinent information, on an industrial scale.
Focused on 'mégadonnées' (Big Data), this technological landscape
must be able to evolve in order to adapt smoothly and quickly:
• to the availability of new data sources: compatibility of internal and
external systems (cloud), migration, Internet of Things (IoT), etc.;
• to different formats of data, models, ontologies and semantics:
structured and unstructured data, regulatory models;
• to the needs for adaptation of data flows: adjustment to operation-
al or decision-making processes, etc.;
• to heterogeneous processing of data: storage capacity, storage
area, aggregation, query, traceability, etc.;
• to varied analysis requirements: depth of analysis over several
years, alignment of various data domains in order to respond to new
questions, etc.;
• to changes in operational contexts and regulatory requirements:
adaptation of models to market changes, predictive models, etc.;
• to the needs for personalised visualisation and communication:
data adapted to the professional context and user profile (patient,
regulatory authority, healthcare personnel, etc.);
• etc.
This landscape is constituted by additional technological compo-
nents (bricks) that each meet a technical or functional requirement.
Through the integration of these components, they represent a tech-
nological landscape that achieves the capacity and processing per-
formance levels necessary for Big Data. The 7 Vs10
therefore define
the horizon of this technological landscape.
10. The 7 Vs are 'Volume, Velocity Variety, Variability, Veracity, Value and Visualisation'.

32
How should the design of IS take into account the func-
tions of knowledge generation?
To define a landscape such as this, the principal subjects to be
considered are the following:
• the availability and granularity of the data essential for the scope of
application sought;
• the data formats to be integrated/displayed;
• the appropriate volumes to be centralised, the acceptable con-
fidence levels for the data to be processed, the requirements for
freshness of these data, and the depth of the data (history);
• the types of processing to be implemented, the need to develop
the algorithms through learning;
• the execution speeds appropriate for the processing operations;
• the flow rates for information to be considered;
• the ability to consume these data on a self-service basis;
• the need for space for prototyping and assessment of new scenar-
ios in agile mode, of the 'Sandbox' or 'Data Lab' type;
• the variety of different tools for Data Mining, retrieval and display;
• the service levels expected by the various consumers of these data;
• the management strategy adopted for the processing of these data
(duplicated, isolated or centralised);
• the good practices made available to meet these requirements
(professional data model, etc.).
And for each of these subjects, we must ask the following questions:
• For whom?
• For what 'professional' objective?

33
A. VARIETY
What the technology provides …
The richness of unstructured data can finally be exploited.
One of the benefits of Big Data is the analysis of unstructured data
(such as video, voice recordings, non-digitised texts, etc.), because
production of these data is more abundant and their use provides
greater added value.
In this context,it is not the 'Big' that is important but rather the 'Smart'.
For example, to understand a document drafted in natural language,
the choice will be based on the technologies and standards of the
'Semantic Web' to perform semantic analysis.
… and what this could mean for prevention
Scientists can now work on new data, irrespective of their original
format.
B. VOLUME
Volume allows exhaustive information.
In 2019, the storage potential of all computers in the world should
be 20 exabytes, namely 1018
bytes.
As a result, the Mayo Clinic has introduced textual analysis tools in order
to have access to a clearer understanding of the free-text notes from
millions of medical records available in its systems. Six types of HL7
messages are now collected in near-real-time, indexed, transformed,
analysed and made available to personnel.

34
Sébastien Verger, Technical Director at EMC, predicted in 2013 that
'the data volumes will increase by a factor of thirty by 2020, reaching
35 zettabytes (namely 1021
bytes) globally.'
Scientists, who are used to the scarcity of information and limited
samples, are scaling up: information is now 'ordinary' rather than
'rare'; 'x%' samples are now 'all', although this sometimes happens
at the cost of the quality of the information.
Exhaustive information means that we can be less exacting about the
accuracy of the information. The data are often in disarray, of variable
quality and taken from a vast number of sources. The information
from unstructured data is often not very dense, but its value lies in
its quantity.
Working with complete data, even if they are imperfect, makes it pos-
sible to gain objectivity in terms of the principle of causality, which is
often a source of errors and incorrect interpretation.
Biology is a domain where a hypothesis is the basis for all reasoning.
The protocol is always the same: an initial hypothesis is stated and its
accuracy or inaccuracy is then verified.
By way of example, rather than stating a hypothesis on the basis
of an observation of a protein, the use of 'Data' makes it possible to
identify a probable trend on the basis of a mass of data on a range
of different molecules. This is what the American Ernest Fraenkel, a
biology researcher at MIT, is working on. He is attempting to con-
struct a unique model that incorporates all types of data: 'My great-
est innovation was to propose a holistic interpretation of the data.'
Ernest Fraenkel is therefore revolutionising biology codes with this new
approach.

35
C. VELOCITY
Associated with Big Data, Data Mining is becoming 'Big Data
Mining'…
The tools for data analysis (statistics and Data Mining) and text
analysis (text-mining) are essential for the use of repositories of
information.
Current Data Mining tools make it possible to analyse numerous
data, but over a sample considered to be representative. These tools
are limited by the associated processing time and/or processing
capacity.
Through the combination of Data Mining tools and Big Data tools
(immense storage capacity, speed of execution of processing), Big
Data Mining makes it possible to process all of the data with pro-
cessing times that have become acceptable and overlapping,
and to move from the descriptive towards the predictive and the
prescriptive.
The ability to combine all of these data for greater transversality and
agility…
With Big Data technologies, data can be factored and aligned in an
analytical ecosystem that offers a level of transversality, making it
possible to respond with greater agility to the challenges facing the
industry.The same data are then consumed, from distinct profession-
al standpoints, by means of Data Mining and business intelligence
tools, by different professional profiles.
Scientists need to generate statistical processing very quickly and to
keep a historical record of those operations.

36
D. THE 'VOLUME/VELOCITY /VALUE' RATIO
A considerable increase in the volumes processed without any loss
of performance
The data exchanged over the internet, originating in some cases from
connected tools, are stored on servers and hard drives. They can be
collected directly. Scalability is the ability of a system to maintain its
functionalities and performance in the event of a significant increase
in workload. But scalability models are not linear and the mere fact of
adding storage capacity does not necessarily improve performance.
The traditional solutions available on the market (IBM, EMC, etc.)
provide the 3 Vs required for Big Data, but each has its own specif-
ic model for the implementation of distributed storage: Cluster File
System or Parallel File System. But these solutions do not provide the
Adverse effects associated with medicinal products cause 10,000
deaths in France each year. In order to create alert procedures based
on past errors, the European PSIP Project (Patient Safety Through Intel-
ligent Procedures in Medication) is proposing, inter alia, that these pro-
cedures be generated on the basis of an automated data search (Data
Mining).
Let’s consider this in terms of the technological landscape. Suppliers
of MRIs have the capacity to ensure a service quality and an availability
rate for their sensitive equipment that is substantially increasing, which
will be of direct benefit for physicians and patients. GE Healthcare relies
on advanced analysis functions and various predictive models applied
to the data sent by its equipment in real time to perform predictive
maintenance. Intervention therefore takes place upstream, before any
service interruption can occur (source: TeraData).

37
same performance or the same degree of scalability in the event of
a ramping-up of operations (when the storage capacity of the disks
increases by 100,000, throughput only increases by 100).
As a result, the 'traditional' decision-making architecture with its da-
tabase is no longer the only reference architecture. It is by thinking
outside the constraints of traditional architectures that the players
involved (primarily the big actors of the internet) have managed to
find solutions. There are currently three complementary reference
architectures to be managed: database, In-Memory and massively
parallel.
Having a website that is always available, being able to collect colos-
sal quantities of information very quickly, dealing with peaks of flows
to that site, ensuring the security and backing-up of the platform –
these are necessities for any website of a health player wishing to
disseminate information to or collect it from medical personnel and/
or patients, while ensuring that the flow of information is secure.
E. VARIABILITY AND VERACITY
The algorithms for 'what'
One of the principles of Big Data is not just to discover correlations
Johnson&Johnson, one of the primary manufacturers of hip and knee
implants, will be working with Watson to create a concierge service for
patients in order to help them prepare better for surgery and support
them during the post-operative period.
The latest generation of pacemakers are smart devices: they send re-
al-time information to the hospital or doctor, making it possible to re-
spond quickly in the case of an anomaly or at least to provide remote
monitoring of the patient's medical status.

38
and trends but to do this without explaining the origin of those phe-
nomena. A correlation is merely the quantification of the statistical
relationship between two values. It will be a strong correlation if one
of the values is very likely to change when the other is modified. Cor-
relations do not provide certainty, merely probabilities. They do not
say why something has occurred, but simply that it has occurred.
Big Data tackles the 'what' and not the 'why'. However, it does so
with huge precision. There is therefore no longer any need to state
hypotheses to be verified: we merely need to let the data speak and
observe the connections between them. In some cases, we will not
even be aware that these connections exist.
However, according to Frank Pasquale from Yale University (The
Black Box Society), neutrality is a myth and the manipulation of
algorithms is a reality.
Based on this observation, the technologies of Big Data are intended
to push this neutrality forward, because they apply to a much broad-
er scope of information, and therefore to provide greater objectivity.
We must, however, be careful to ensure that we always correctly
qualify, contextualise and provide perspective for the data, because
the degree of responsibility in the field of health is high.
In 2009, in the midst of the H1N1 flu pandemic, the American Min-
istry of Health asked Google for help. By locating the map position
and origin of keywords typed in the well-known search engine,
the engineers practically succeeded in tracing the course of the
Last December, Google announced the creation of Verily, a new subsid-
iary bringing together its projects in the medical sphere and based on
the former entity Google Life Science. Equipped with hardware, soft-
ware, clinical and scientific teams, Verily works on platforms, products
and algorithms intended to identify the root causes of diseases and
to analyse the most appropriate treatments in order to provide better
diagnosis and care.

39
epidemic, with a few minor deviations: the word 'flu' in the search
engine reflected a simple concern on the part of the internet
community about the approach of winter.
F. VISUALISATION
Visualisation for interactions
Big Data technologies make it possible to consume information
completely independently, to have access to maximum interactivity
and to obtain instantaneous responses.
Visualisation can make it possible to detect invisible trends immedi-
ately, and to highlight the most convincing data.
Visualisation is not just a method of representation, but also a
means of encouraging the stakeholders to accept the data, study
them, share them and better understand them. A tool to support
understanding and decision-making, it can become a method of
communication.
This complex problem of data visualisation is the core of a programme
designed to explore the culture of health driven by an American foun-
dation. For example, which method of visualisation is the most appropri-
ate to communicate about the importance of preventing cardiovascular
disease? These graphic designers have no understanding of prevention
and have thus worked on a list of 16 scenarios for the communication of
medical risks. At the end of the chain, citizens and patients judged the
various proposals.

40
G. CURRENT DIFFICULTIES
In an technological context that is still relatively immature:
• difficulty in expressing the requester's requirements to the
provider;
• difficulty for the provider in understanding the requester's
medium- and long-term roadmap;
• difficulty for the requester in identifying the first structuring pro-
ject and the successive steps leading to the medium- or long-term
roadmap;
• underestimation of the efforts required to align and clean up the
data;
• multitude of technologies and technological combinations;
• emerging expertise of integrators;
• absence of objectivity about the industrial capabilities of open-
source platforms in terms of serving operational requirements and
professional requirements;
• underestimation of the requirements for data governance;
• difficulty in obtaining a sponsor for general management that is
able to drive a transverse project of this kind.
H. CONCLUSION
• The functional possibilities provided by Big Data are immense.
• The regulations are still open to proposals for technological
innovation.
• The principal challenges are still organisational or ethical.
I. A UNIFYING INFRASTRUCTURE –
ESSENTIAL FOR BIG DATA
Managing such a rich and complex environment requires a new

41
approach by the technological landscape.
Therefore, we talk of an 'ecosystem' that is able to process the data
effectively in all of its expressions and dimensions, to support the use
of innovative tools for visualisation and analysis, to optimise the data
integration process and to adapt immediately and without compro-
mise to the various constraints (technical, operational and regulato-
ry).
The parallel and close-in processing of data (In-Database) are essen-
tial and represent a vital characteristic of the Big Data infrastructure.
Processing operations are performed in parallel, a unique guarantee
of response time and service level commitment that is essential for
serving users whose requirements increase with the expansion of the
data available.
The intrinsic industrial capabilities and flexibility of a platform of this
kind therefore enable the combined use of a range of data process-
ing technologies. This modular, agile and coherent system is de-
scribed perfectly by analysts such as Gartner using the term 'Logical
Data Warehouse Architecture Model'.
This architecture model allows organisations to access the data,
refine them and retrieve them in a way that is reliable while at the
same time guaranteeing the availability of the system, and the
performance, concurrency and variety of processing operations.
Selecting the appropriate tool to address the problem identified is
a key principle of good engineering practice. On this fundamental
principle, the concept of a Big Data ecosystem is intended to pro-
vide the right technology to perform the right processing. The adage
'a good workman chooses the right tools for the job' applies perfect-
ly. There is not currently one technology that is capable of resolving
all problems.
This agile architecture can be broken down into the following five
components:
1) the Data Lake;
2) Data Discovery;
3) an Integrated Data Warehouse;

42
4) the Data Integration Layer;
5) a proven resource allocation engine (Integrated Workload
Management).
This architecture is fully integrated and interoperable.
The key is to understand which parts of an analytical workload should
be allocated to each of the technologies within the architecture and
to orchestrate the interaction of these various components. The ana-
lytical workload can involve several sources and types of data, requir-
ing a dispersal of processing operations over different technologies
within the ecosystem.
For example, all of the components of this architecture will be need-
ed for the correlation of behavioural data originating from the mo-
bile application providing the monitoring of patients with type 2 di-
abetes (activities, recording of vital signs, glucose level, well-being
assessment, meals and calories, etc.), with data from the community
website, prescription data for a patient population and the individual
medical file. The Data Lake will be needed more for the integration
of data from smartphones or the community website, while Data Dis-
covery will rely on the Data Integration Layer to ensure transparent
access to the data in the Data Lake and the Integrated Data Ware-
house. Data Discovery will then apply the various algorithms for
the analysis of behaviours, and lastly the Integrated Data Ware-
house will be used for more structured data based on prescription
or the patient file, or for example the recovery of the results from
the Discovery phase in order to operationalise these results vis-à-vis
patients.
The various stages
of Big Data processing.
Source: CXP 2015
EXTRACT
Odata, ETL,
Webcrawler Data cleaning,
Data Quality
STORE
Hadoop,SGBDR,
Cloud
RETRIEVE
Ad-Hoc Query Reporting,
Scoreboards,
Data Visualization
ANALYSE
Datamining, «R», Analytics,
Data Discovery, Machine
Learning
SHARE
Portals, Search,
Collaboration
Self-Service
INTEGRATED
SUITE

43
Data Lake
The Data Lake component houses the raw material: unrefined raw
data with a limited information density, entered without a strict
organisational structure so that it can then be transformed and
correlated in order to provide new perspectives and professional
value.
The continuous capture of all data is an implicit goal to be achieved,
because any data ignored and therefore not used will never be able
to reveal their potential. The implementation of a successful Data
Lake must meet the following key characteristics:
• scalability;
• low cost of the stored data per terabyte;
• support for structured, semi-structured and unstructured data;
• openness to external solutions and tools.
Data Discovery
The standard use for the Data Discovery function falls within R&D
activities undertaken by data scientists and business analysts as part
of an analytical approach based on exploration and statistical mod-
elling. It is unique in its enormous agility to correlate data originating
from the Data Lake, Integrated Data Warehouse or external sources,
and in its quick iteration in order to find and rapidly reveal hidden
facts and proven correlations. It also offers business users the option
of using these results in the form of reusable applications.
The implementation of a successful Data Discovery solution must
meet the following key characteristics:
• The vast range of available analytical techniques (statistical and tex-
tual analyses, analysis of graphs, path analysis, analysis of models,
etc.) and script execution techniques (R, SAS, SQL, MapReduce, etc.).
• The effectiveness of the process of data acquisition, preparation,
analysis and visualisation.
• Support for structured, semi-structured and unstructured data;

44
• Openness to external solutions and tools.
Integrated Data Warehouse
The Integrated Data Warehouse component integrates and deploys
the data as a finished product for business users for operational,
decision-making purposes. Its role is to unify and bring together a
single version of the data, to be used for multiple purposes (mul-
ti-domain management) to meet varied user requirements. The de-
gree of transversality sought will be all the more effective where the
information system is actually integrated as a whole.
The requirements associated with multi-domain management mean
that the solution must be able to provide constant performance
demonstrated by contractual service level commitments vis-à-vis
end users.
The solution must also make it possible to mix production data with
non-certified data within analytical Data Labs or Sandbox. In other
words, self-service Data Labs enabling users to create a space for
prototyping and evaluation in which they can test, verify and validate
hypotheses by combining certified production data (read-only) with
new sources of information (flat file, real-world data, data from the
Data Lake or Data Discovery).
The implementation of a successful Integrated Data Warehouse
must meet the following key characteristics:
• a high-velocity solution built on massively parallel processing
technology;
• robustness and high availability of the solution;
• linear scalability in line with the change in requirements and
constraints;
• analytical functions performed as close to the data as possible
(in-Database);
• support for structured, semi-structured and unstructured data;
• advanced logging, traceability, auditability and security functions;
• flexibility of self-service spaces (Data Lab or Sandbox);

45
• openness to external solutions and tools;
• the ability of the solution to absorb operational constraints simul-
taneously (volume, number of simultaneous queries, performance).
Simultaneous development over several dimensions
Integration Data Layer
To be productive, users do however need technologies that free
them of the need to transfer data from system to system, or even to
understand the mechanisms by which data are moved.
This component is particularly important at a time when open-
source technologies are becoming a major component of Big Data
environments.
It is true that different technologies originating from different sup-
pliers are being used together within analytical and operational
systems. These non-integrated technologies are often considered
by users to be distinct systems. It is therefore difficult to exchange
data and to perform processing operations between separate
components.
1
2
3
4
5
6
7
8
VOLUME
OF DATA
NUMBER OF
CONCURRENT
QUERIES
COMPLEXITY
OF QUERIES
(NUMBER OF
JOINS)
COMPLEXITY
OF THE DATA
MODEL
VOLUMES HANDLED
BY QUERIES
QUERY FREEDOM
FRESHNESS
OF THE DATA
MIXED
LOADS
MB
GB
TB
simple
star
Integrated
multiple
stars
BFN
5 15 30
50
500
1000
5 TB
20 TB
100 TB
Operational
Decisional
Batch
input
Minute Hour Day
Pre-programmed
reporting
Ad hoc,
ROLAP
Data
mining

46
The Integration Data Layer provides a layer of interoperability and
therefore transparent two-way interaction for data and for analytical
processing operations between the various components of the Big
Data ecosystem. It takes care of managing the performance of que-
ries that use several analytical engines and data warehouses (Data
Lake, Data Warehouse, etc.), in a manner that is optimised and se-
cure, thus integrating multiple systems so that they create a single
unified system.
The implementation of a successful Integrated Data Layer must meet
the following key characteristics:
• two-way processing;
• optimisation function during the performance of queries on multi-
ple systems and heterogeneous technological environments;
• integration with open-source environments;
• possibility to query the ecosystem from any component (Data
Lake to Integrated Data Warehouse, Data Discovery to Data Lake,
Integrated Data Warehouse to Data Lake, etc.) ;
• limited data movements or data duplication.
Integrated Workload Management
Effective workload management is crucial in providing corporate
users with responses to their questions in real time and meeting
performance and service level commitments for Big Data infrastruc-
tures. The Integrated Workload Management component deals with
dynamic resource allocation. Its role is to simplify and automate data
management, optimise the allocation of machine resources and
manage the coexistence of very different uses on the same platform.
This means that each task can be performed under optimal condi-
tions, guaranteeing the various service levels in terms of operational
variations.
Through 'deterministic prioritisation', the engine ensures that the
most important tasks are performed first. Its mechanisms also make
it possible to stabilise response times and to optimise the use of
resources during peaks, whether planned or unplanned.

47
The implementation of a successful Integrated Workload
Management system must meet the following key characteristics:
• holistic vision of the Big Data ecosystem (Data Lake, Data Discovery
and Integrated Data Warehouse);
• dynamic changing of priorities without halting queries or the
system;
• maturity of the dynamic resource allocation engine;
• hierarchised task allocation (priorities);
• filter, exception management and planning functions;
• definition of allocation criteria (rules);
• possibility of allocation of service levels;
• integration into an integrated surveillance tool.
The GLIDE project (Global Integrated Drug Development
Environment), implemented in 2015 within the Roche pharmaceuti-
cal group, combines a vast range of internal and external data within
a single ecosystem (clinical studies/new processing data – gener-
ally SAS Datasets, laboratory data – blood, x-rays, electroenceph-
alograms (EEG); medical data, genetic data, biomarkers or even
real-world data).
This analytical ecosystem has the effect of significantly reducing
processing time (from several days to some hours or minutes), but
also of enabling R&D teams to develop better targeted therapies,
reduce the adverse effects of new molecules or even, using all of this
historical data of observations, reposition drugs for new therapeutic
purposes.
J. GLOSSARY
Unstructured data: texts originating from the processing of texts or
content of electronic messages, audio files, voice files, fixed images
or videos.

48
Structured data: spreadsheet pages organised in tables, databases
where the data are organised in connected tables.
Big Data: literally 'big data'. English expression describing all data
that are so big that they become difficult to process using traditional
database management tools. Within the HDI, the term has a broader
meaning and describes all data irrespective of source, format and
volume.
Smart Data: focuses on the data relevant in relation to their specific
objectives.
Data Mining: detection of information in a database.
Tools capable of detecting the information hidden 'at the deepest
levels' of the 'data mine'. This does not concern database query sys-
tems, spreadsheets, statistical systems or even traditional data anal-
ysis systems.
Data Mining is undertaken using several different approaches:
• 'Verification' approach: the user has an impression or general idea
of the type of information that can be obtained from the data. The
database can then be used to 'quantify' that impression. It is clear
that the data extracted, and the decisions made using those data, de-
pend exclusively on the user's impression of the important parame-
ters of the problem (age, geography, etc.), an impression that is often
correct but not exhaustive.
• 'Discovery' approach (Advanced Data Mining) or search for hid-
den information: the user understands that the quantity of data to
which he or she has access is considerable and thus the optimal and
exhaustive detection of important structures or relationships is com-
pletely beyond the reach of human users. Users must therefore use
advanced data analysis methods to detect the hidden information
(which may be the most interesting).
Text-mining: all of the methods, techniques and tools for making use
of unstructured documents. Text-mining relies on linguistic analysis
techniques.
Architecture: describes the general structure inherent in an IT sys-
tem, the organisation of the various elements of the system (software

49
and/or hardware and/or human and/or information) and the rela-
tionships between them.
The following list includes some different architectures:
• Traditional decision-making architecture (Oracle, SQL Server,
MySQL, Informatica, Datastage, etc.): this architecture performs well
where the volume of data to be transferred between each stage is
limited.
• In-Memory architecture (Qlikview, ActivePivot, HANA, etc.): this
architecture can offer high-performance analysis services, or indeed
real-time services (updates to data and recalculation in line with clus-
ters) and simulation services.
• Massively parallel architecture (Hadoop, TeraData): this architec-
ture can store an immense quantity of data (unlimited) elastically.
Scalability: ability of a system to function correctly with increasing
workloads.
2. 'KNOWLEDGE BY DESIGN'
OR SEMANTICS AS THE KEY TO VALUE
The information systems within the health sector should in all cases
be able to contribute to public health, to support our health system
and to generate new knowledge. This is only possible if we define
a certain number of principles and rules that create a coherent,
stable and multidimensional technological, legal, financial, organi-
sational, institutional and political framework. Within this 'planning
and development' approach, the production of data, their type and
the conditions under which they can be reused represent a major
investment for the future. The creation of value expected as a result
of the digitisation of the sector and the large-scale production of data
therefore require a strong commitment in terms of interoperability.

50
Semantics, a source of infinite value creation:
'knowledge by design'
The deployment of internet infrastructures and the organisation of
the various communication protocols benefit all business sectors.
The specific issues affecting the health sector cannot be found here,
but rather in the more 'business-specific' layers, first among which
is semantic interoperability, the principal source of value with the
emergence of the Internet of Things. And for health, this constitutes
a complex lever to be grasped and understood, since it carries enor-
mous promise. Producing electronic information certainly provides
direct savings in terms of time and money. But ensuring that these
data make sense and can be interpreted, analysed and reused cre-
ates opportunities that are simply infinite. Semantic repositories are
the printing press of the 21st
century and will make it possible to pro-
duce and transmit the bulk of knowledge over the coming centuries.
Data production is experiencing considerable growth and this pro-
duction must therefore have access to such repositories as quickly
as possible, because we will not be able to turn back the clock on
those data. The challenge of being able to automate some or all of
the processing of these data taken from multiple sources and pro-
duced in contexts and for purposes that are different in each case
is becoming an extremely urgent imperative. The deployment of
an IT structure that continues, as is still often the case today, to use
electronic data that are unstructured and not reusable must be re-
placed as soon as possible by the dissemination of semantic repos-
itories instead of, for example, unusable JPEG documents. In more
macroeconomic terms, the capacity to reuse data will help to share
the costs of production and enhance the creation of value. The right
to reuse data for purposes referred to as 'compatible' introduced
by the new European Regulation is a measure that has substantial
impact. The model that until now required that data be generated
specifically for a dedicated and most often unique purpose should
now be reserved for ad hoc studies making use of concepts that are
insufficiently widespread to enable the use of a common vocabulary.
The semantic integration of different systems constitutes a major
challenge that must be taken into account far upstream, in order to
enable the application of a 'translational' approach establishing a

51
permanent bridge between treatment activities on the one hand and
research on the other. This concept of knowledge by design, by anal-
ogy with privacy by design, promoted in the field of security, must
ensure that the bulk of the data produced has a common meaning
that can benefit public health in particular.
Other countries are concerned with this issue, as we can see from
the creation of the CDISC consortium (Clinical Data Interchange
Standards Consortium) by the American Food and Drug Adminis-
tration. This agency has the task of sharing clinical trial data around
common, free-access semantic repositories. We should also con-
sider the strategic linguistic issue. The 70 countries making up
the French-speaking world and the 220 million French speakers
worldwide have an interest in making sure that English-speaking
codifications are not the only such systems in general use.
Each item of health data can potentially contribute to prevention,
provided that its meaning is clear and that the reuse of the data is
taken into consideration when the system is designed.

52
ARTICLE IV
THE DEFINITION OF A NEW
PARADIGM FOR STAKEHOLDERS
1. BIG DATA: A NEW PARADIGM
FOR THE PHARMACEUTICAL INDUSTRY
The pharmaceutical industry understands today that the quantity
of available health data, combined with current analysis capacities
and algorithms, constitutes a major factor in the emergence of new
practices in epidemiology, personalised medicine and prevention,
and for research and development of new health services that will
transform patient care and management.
A large number of pharmaceutical laboratories are developing med-
ical products intended for patients suffering from chronic diseases,
which require long-term monitoring. A drug alone is no longer seen
as an isolated treatment variable. It forms part of a broader treat-
ment scenario that has a decisive impact on long-term treatment
compliance by the patient.
The drug is now just one element among others making up integrat-
ed health solutions that form part of the 'beyond the pill' strategy.
To ensure the efficacy of a drug, the pharmaceutical industry needs
to have access to tools to measure the results: improvement in clin-
ical results, and the prevention of the adverse effects and serious
events that may occur in patients. In a patient-centred model, statis-
tics play a vital role. They contribute to understanding the behaviour
of the patient and thus make it possible to identify the factors that
determine treatment compliance by the patient11
.
Pharmaceutical laboratories are moving in this direction by devel-
oping and providing innovative health services for patients: not just
11. Interview with Pascale Witz, Healthcare Data Institute Newsletter, June 2015.

53
medicinal products, but also the accompanying healthcare solutions.
Many of these solutions involve the collection and analysis of data
in real time using appropriate algorithms, so that quick treatment
decisions can be made.
The technology will change the patient experience (by improving
care and cooperation by the patient) and, by enabling the analy-
sis of large quantities of data, will make it possible to initiate a new
management and care process. Improved clinical results for pa-
tients is the objective to be achieved in order to consolidate a suit-
able business model. As a consequence, the technology combined
with therapeutic monitoring will have greater impact compared to
technology alone (see Diagram 1).
Diagram 1: The technology of connected objects is changing the patient
experience, and the collection and analysis of Big Data are facilitating the
process. The diagram shows an example of an integrated treatment approach
for monitoring and supporting patients suffering from respiratory problems.
Several innovative software systems for health data have already
moved on from the stage of retroactive reporting (data collection)
to incorporate predictive functionalities (data analysis), making it
possible to warn patients of a potential adverse effect.
At the same time, recent technological advances have facilitated
the collection and analysis of information originating from multi-
ple sources; a major advantage in the area of health, given that the
BIG DATA
• Population management:
Focusing HCP workﬂow
on the problem patients
• Data collection:
Integrating data from
public sources, partners
and patients
• Data analytics:
Segmenting the patients
as consumers - predicting
behaviors
• Data-driven care:
Directing individualized
services, motivation and
support to each patient
DRUG DELIVERY
Hassle-free
MEDICAL MONITORING
Mindless
DIAGNOSIS/PREVENTION
Engage early
MOTIVATION & SUPPORT
Personalized and
engaging
ROUTING MONITORING
At home or "as convenient
as your nearest pharmacy"
MEDICAL VISITS
When I need it
PATIENT

54
personal data relating to a patient can come from various measure-
ment technologies, in many cases with little or no active participa-
tion from the patient. For example, portable ingestible or implant-
able sensors and portable digital applications are able to collect
and transmit data independently; the analysis of data essentially in
real-time will enable healthcare professionals to anticipate and avoid
imminent crises in patients.
A. PATIENT MONITORING AND INDIVIDUAL
SECONDARY PREVENTION – CHRONIC
DISEASES AND DIGITAL HEALTH
The value of prevention covers both prospective prevention of pop-
ulations and individual secondary prevention, the latter being spe-
cifically intended to avoid or anticipate serious events, relapses and
crises in patients suffering from chronic diseases (such as diabetes,
asthma, rheumatoid arthritis and cardiovascular disease).
The implementation of individual secondary prevention generally
requires integrated health solutions that include medical devices,
portable sensors, applications for smartphones, the collection and
analysis of health data in real time using appropriate algorithms, and
real-time feedback and notification of patients and healthcare pro-
fessionals, in order to enable interventions in emergency cases (see
Diagram 2).
Intelligent devices (such as wireless blood pressure monitoring, de-
vices used for passive and active data collection, wristband sensors
for monitoring sleep quality) are the common components of ad-
vanced platforms for patient surveillance and participation, which
are designed to improve patient treatment and management, and
therefore to achieve better clinical results and reduce the total cost
of treatment.

55
Diagram 2: Connected objects, sensors, software and data analysis are the necessary
components of integrated health solutions.
Doctors and patients are beginning to use digital tools that enable
more effective management of health. In the future, pharmaceutical
laboratories and other stakeholders will adopt the existing technolo-
gies and develop new ones. As part of that process, they will create a
more connected digital health ecosystem that will benefit everyone:
• The pharmaceutical laboratories are creating applications and
other platforms to enable better management of diseases and treat-
ments, providing information about illnesses, drugs, appropriate
doses, expected clinical results, health objectives and healthy life
choices.
• When a doctor prescribes a drug, he or she will also prescribe the
digital application.
• Patients can also collect and monitor their own data in real time.
They can send them to their GPs for in-depth assessment, and make
adjustments themselves to their drug treatments or lifestyles.
• Patients can also send their data to pharmaceutical laboratories,
which may then use this information to develop treatments and im-
proved health services targeted more on patients, and thus conduct
corresponding research.
PATIENT PHYSICIAN
DISEASE
MANAGEMENT
APPS
PATIENTS
& PHYSICIAN
SOFTWARE
CONNECTED DEVICE
INTEGRATED
SOLUTIONS
CONNECTED DEVICE

56
B. EXAMPLES OF INITIATIVES FOR INTE-
GRATED HEALTH SOLUTIONS USING
REAL-TIME DATA ANALYSIS
Ginger.io is offering a mobile application in which patients suffering
from specific diseases can ask, in cooperation with doctors, to be
monitored by means of their mobile telephones and supported in
the form of behavioural health treatments. The application records
the data on calls, text messages, geographical location, and even
physical activity. Patients also respond to questionnaires provided
on their mobile phones. The Ginger.io application integrates pa-
tient data with public research on behavioural health by the National
Institutes of Health and other sources. The information obtained can
show, for example, a lack of movement or other activity that would
indicate that the patient does not feel physically well, and that there
are irregular sleep cycles (revealed by late calls or text messages)
that could indicate that an anxiety attack is imminent12
.
Diabetes
There is no absolute rule as to the dose of insulin to be injected.
When a patient begins treatment for diabetes, he or she must move
through steps (dose-ranging). The challenge for diabetic patients
is to maintain the levels of sugar in the blood. Hypoglycaemia (un-
derdosing) or its opposite hyperglycaemia (overdosing) can be the
source of serious problems.
A solid treatment plan will include a solution for the continuous mon-
itoring of glucose levels, based on data from connected medical
devices and analytical data capabilities, in order to prevent a hypo-
glycaemic episode well before the clinical symptoms manifest. The
Diabeo solution from Sanofi is one example of an advanced tech-
nology providing a telemedicine solution enabling the individual-
ised adjustment of insulin dose, based on glycaemic measurements,
patient treatment regimes and physical activity.
12. Unlocking the full potential of data analytics for the benefit of all, Healthcare Data
Institute 2015.

57
Mobile monitoring of vital signs for diabetic patients using portable
(wristband) connected objects makes it possible to share data in al-
most real time with the patient's healthcare professionals, and thus
anticipate the risk of future complications, including diabetic foot
and ocular conditions.
An automatic portable bionic pancreas using a sensor connected
to a smartphone and with algorithmic software has been shown
to perform better than traditional insulin pumps, according to an
article published in the New England Journal of Medicine, July 2014.
Interesting White Papers
• Predictive Medicine Depends on Analytics, Harvard Business School
• The Big Data revolution in healthcare, McKinsey & Company
2. THE INSURANCE SECTOR AND BIG DATA:
PREDICTIVE AND THEREFORE PERSONALISED
HEALTHCARE?
A. BACKGROUND
We live in an era of revolution and technological change defined
as the third industrial revolution. The source of this revolution is the
datum, by analogy sometimes compared to what was represented
previously by coal, and then oil, in their transformations leading to
the mass production that is always at the heart of our current society.
Crude oil, like raw data, is nothing in itself. Refined and transported,
it created the petrochemical revolution and the current transport rev-
olution. Placed correctly in context and analysed using algorithms
that benefit from the constant increase in growing computing power,
raw data will lead to an incredible creation of value.
This revolution is based on the capacities of Big Data. These capac-
ities are based first and foremost on an exponential increase in the
volume (V) of data available due to the increase (V for Variety) in

58
connected objects (smartphones, PCs, watches, consoles) and sen-
sors (altimetry, home automation, vehicles, household appliances,
etc.). All of these increasing interconnections define the concept of
the Internet of Things (IoT) or the Internet of Everything (IoE), which
generates a doubling of the volume (V for Volume) of the data availa-
ble every three years. Some people are calling this Internet of Things
the new steam engine of the 21st
century.
The consequence of this 'breaking wave' is the 'datafication' of the
world or the digitisation of the real. We should note that the anal-
ogy with coal or oil, as finite resources, stops there. Data are an
inexhaustible resource.
Big Data is also based on the current power (V for Velocity) of stor-
age capacity and calculation capacity, and on the sophistication of
the algorithms.
For the first time in the history of humanity, we have access to a vol-
ume of data and a computer power that makes it possible to escape
the constraints of sampling, accuracy of measurement or the need
to have to work on limited data analysed using standard statistical
calculation.
While this data revolution will greatly affect all economic organisa-
tions and sectors, three sectors are particularly impacted: Housing
(home automation, energy saving, household appliances), mobility
(connected and independent vehicles) and health (efficiency of care,
ageing, monitoring of chronic diseases).
These sectors are the primary markets for insurers and they play a
major role in them.
Where these markets meet, insurers will therefore be compelled, by
the Internet of Things and Big Data, to rethink their service offers,
internal organisations and business models. This will mean, on the
one hand, that they will not be left behind and, on the other hand,
that they will be in a position to exploit the extraordinary opportunity
to create value (the fourth V of Big Data after Volume, Variety and
Velocity) presented to them.

59
B. IMPACTS AND PERSPECTIVES
Insurers are historically collectors of data which they then use for
analysis. These analyses are based on claims made, on a vast set
of societal data and on a pooling of risk enabling the calculation of
premiums.
Until now, the analysis and assessment of risk have paid little re-
gard to the use made of the insured item, whether this was tangible
(property and casualty insurance) or intangible such as health (health
insurance). The new techniques for the collection of data, the more
widespread use of connected sensors, the digitisation of health with
its increasing flow of associated data and the introduction of ad hoc
algorithms able to adapt in order to follow the changes in the be-
haviour of an insured person will shatter the traditional model of
actuarial calculation.
For insurers, the issue is to bring together the aggregated histor-
ical knowledge they have of their insured people and claims (of-
fline data) with the data from the new digital environment (online
data) that are now accessible (visiting of websites, activity on social
networks, home automation, on-board units, connected devices,
etc.). This merger of the offline and online enables targeting by the
analysis of weak signals and predictive, personalised monitoring of
insured persons.
This new use of data impacts the entire insurance value chain,
namely:
• design of products and pricing;
• knowledge of the customer and the market;
• prevention of fraud;
• management of claims;
• new services.
One preliminary statement should be noted: in France, the storage
and use of these personal data are governed by very strict regu-
latory requirements, particularly in relation to health data, which

60
significantly restricts the possible avenues open to insurers.
The design of products, pricing,
knowledge of the customer and the markets
For the design of the products, and to refine understanding of the
customers and the market, marketing has or will have access to in-
creasingly high-performance tools intended to improve segmenta-
tion, personalisation and identification of insured persons.
As explained above, the increasing number of sensors, the reconcili-
ation and cross-referencing of data, and the power of the algorithms
mean that we can now no longer look directly and only at the value
of the number of claims but also at the use and behaviours observed
in relation to the scope insured.
There are countless examples and possibilities. Offers are already
appearing that make it possible to differentiate certain behaviours
of drivers in terms of premium. In the USA, certain health insurance
negotiations depend on the results identified by connected objects.
Similarly, the analysis of weak signals from the IoT makes it possible,
for example, to reduce the risk of attrition (loss of an insured person)
that can occur during a change of vehicle (detection of unusual que-
ries on websites with vehicle offers) or housing (surfing on property
sites) and anticipate this loss by proposing tailored offers.
Pricing will benefit from all the power of the new tools. The usual var-
iables (sex, age, type of vehicle or housing, etc.) are being enriched
by behavioural and usage data that enable a segmentation of the
product offer and real-time adaptation of premiums.
Insurers are moving from a descriptive view of the events on which
actuarial calculations are based to a predictive view that makes it
possible to adjust the management of risk as closely as possible to
the person.

61
Fraud prevention
This issue is vital for insurers. Close to 20% of fraud cases are not de-
tected. The cost of these cases weighs heavily on the claims process
and therefore on the calculation of premiums. Through its analysis of
data and weak signals, and through its ability to detect unusual be-
haviours and inappropriate usages, Big Data promises a potential
significant drop in this cost. Quite apart from an associated return
on investment, the margin provided suggests there could be a fall in
contributions and/or the design of new offers.
Management of claims
In addition to the reduction in the number of claims expected as a
result of the monitoring and personalised knowledge mentioned pre-
viously, the increasing digitisation of process and flows, and the result-
ing internal restructuring, will enable an optimisation of the manage-
ment of claims, which adds to potential productivity gains.
We should note that with regard to France, the implementation of the
health insurance card (Carte Vitale), combined with remote transmis-
sion from local health insurance funds (CPAM) to top-up health insur-
ance agencies, has already broadly permitted this optimisation.
New services
These new services are generated by the new marketing tools that
identify new offers and requirements resulting from predictive, per-
sonalised monitoring. These new services are improving CRM, in-
creasing customer loyalty, reducing attrition, and promoting a
reduction in the cost of recruitment campaigns. They are also contrib-
uting significantly to a strategy of differentiation and the search for a
competitive advantage.
In short, insurers can expect that the Internet of Things, connected
tools and Big Data will provide them with increased productivity and
improved operational quality. Above all, they can refine their prod-
uct offers and their services for insured persons (gains in well-being,
reduced or adjusted premiums, etc.).

62
In this regard, French legislation, which provides considerable protec-
tion for consumers (in particular the national data protection agency,
the CNIL), could become an asset by making it possible to organise
the collection and provision of all necessary data, and establish a rela-
tionship of trust between insurers and their insured customers.
These insured customers must retain control of their data and must
be informed (information in customer spaces on the insurers' web-
sites, communication push and targeted information campaigns)
and convinced about the uses and purposes for which their data are
collected.
Transparency makes it possible to establish trust, the trust to authorise
collection. Collection makes it possible to achieve the critical volume
that enables analysis.
Analyses lead to personalised or predictive diagnosis using algo-
rithms that embody the expertise and industrial secrets of the insurer.
These algorithms will become the core of the value added by an insur-
er and the guarantee of its performance in the market, in strict compli-
ance with the legislative context and an ethical approach that favours
the individual in all situations.
Insurance and health
In the health sphere, we should note that there is pressure on health
insurers. In most developed nations, given the increasing scale of
chronic diseases, healthcare expenditures are growing faster than
GDP. In France, this cost is above 11% and we have seen this amount
essentially triple over the last three decades.
This increase can be explained by the increasing age of the popula-
tion (as there are therefore more chronic illnesses to be treated over
longer periods), by the high cost of new treatments (new oncology
molecules) and by the new techniques for examination and surgery
(imaging, catheterisation, transplants).
The emergence in daily practice of nanotechnologies and uses of
genome technology are unlikely to reverse this trend.
These continually growing costs are essentially consumed by curative

63
treatments and leave only a very limited amount for prevention (about
2%).
As a consequence, pressure is growing on the top-up portion of the
market devolved to health insurers. This tension can also be seen in
the areas of dependency, disability and ageing.
This trend is also apparent in countries where the role of private
health insurance is more significant than it is in France. The problem
in this situation will then be to keep the amount of the premium at an
acceptable level for the payer.
Health insurers are therefore seeking new business models that can
address these difficulties. In this context, the Internet of Things and
Big Data, because of their capacities in relation to prediction and indi-
vidualisation of risk, are seen by the sector as new tools that can gen-
erate value while adhering to the traditional pooling model. Indeed,
more specific pricing of the risk for an individual within a group does
not prevent the pooling of risk (within a company, a field, a sector or
even on the basis of principles associated with non-discrimination),
and, conversely, new, more efficient forms of pooling should appear.
Furthermore, top-up insurers, through their role as financiers, must be
able to become more important players in the patient/care system
interaction, as a better understanding of insured individuals will be
reflected by the provision of services or options that are even better
suited to the needs of those individuals.
It is therefore necessary to acknowledge that, as part of the digital
economy, the progressive and growing introduction of ICT and NBIC
will result in a very deep-seated disruption to this interaction: move-
ment of treatment paradigms towards prevention and well-being,
changes in the way the health system is organised, new behaviour of
patients (sometimes identified as consumer-players) who are involved
and connected (social networks, websites, etc.), modification of the
patient-healthcare professional relationship, etc.
In summary, the digitisation and 'datafication' of health are creating
new forms:
• of the coordination of patient monitoring (platform of medical and
welfare services, websites, etc.);

64
• cooperation between healthcare professionals by means of in-depth
interaction and permanent connection;
• management by the patient, including when mobile, of associat-
ed health information (storage, sharing, summary, monitoring, alert,
transmission, etc.);
• sharing and distribution of knowledge between players, including
insurers and any new entrants (GAFA).
These changes are or will be reflected by:
• the optimisation of the care path;
• a reduction in hospitalisations;
• an improvement in treatment compliance;
• a better allocation of resources intended for prevention.
Health insurers, as financiers of risk and financiers of the system, can
only become involved and support these trends driven by the Internet
of Things and the tools of Big Data.
At the same time, the concept of health and the way in which it is man-
aged are changing and expanding. Indeed, on the basis of the well-
known definition from the WHO (a state of complete physical, men-
tal, and social well-being, etc.), a change is taking place, with a shift
from the the historical purely biomedical concept towards a holistic
approach that attempts to understand all of the needs of a person
(affective, health, nutritional, cultural, etc.).The determinants for health
are becoming broad and multiple (behaviour, environment, etc.).
We must therefore learn to manage health as a rare resource on the
basis of this holistic approach, which will give priority to prevention
rather than cure.
The promotion of health thus put in place fosters the development of
prevention and, in particular, the notion of well-being. It is in these two
sectors (prevention and well-being) that the connected sensors and
Big Data analyses are particularly pertinent and powerful.
This avenue (management of health capital to thus reduce health
risk, extensive use of smart sensors and Big Data) is particularly iden-
tified and favoured by insurers, who are expecting positive impacts

65
throughout their value chain and namely the following:
• A possible financial saving (monitoring of chronic diseases as part
of secondary or tertiary prevention, decrease in the number of claims
through an improvement in hospitalisation days, improvements in
drug compliance, a fall in the undue cost of avoidable services, etc.).
• A strengthening of the customer relationship (better known, and
therefore better monitored).
• The possibility of new services (decrease in the rate of attrition, bet-
ter customer loyalty, policy of differentiation, search for comparative
advantages). This is one example: monitoring of admission/discharge
from hospital, prevention of musculoskeletal problems, prevention
of work-related stress, nutritional coaching or coaching in physical
activity, etc.
• An adjustment to premiums, increased productivity, reflected in an
improvement in operational quality.
The culmination of this trend initiated by smart tools and the 'datafica-
tion' of health will accelerate through the arrival of nanotechnologies
but above all through the increasing role of genomics.
The analysis and decoding of the human genome are being auto-
mated to an increasing extent, and the significant cost reduction will
enable greater use of this technology in treatment. The power of
the technologies implemented by Big Data makes it possible to de-
tect areas of high risk potential and to optimise the therapeutic and
insurance response.
This genome revolution is completing a fundamental paradigm shift,
namely the switch from standardised medicine (the same treatment
for a given condition irrespective of the population affected) towards
personalised medicine (one illness, one patient, one environment,
one treatment), also referred to as precision medicine. It is in this new
landscape that the insurance business of tomorrow is developing.
The question of trust, already mentioned above, is one that is becom-
ing particularly sensitive. Of course insurers must comply with the laws
and legislative frameworks relating to health data and the protection
of privacy.

66
They must also demonstrate transparency and know how to convince
insured individuals of the relevance of their approach and the positive
benefit they can expect from the collection and analysis of all the data
gathered.
Public authorities must also move our current legal system forward in
order to protect both personal freedoms and patient rights and the
competitiveness of French insurers compared to their international
competitors.

67
ARTICLE V
ANONYMISED DATA, PSEUDONYMISED DATA:
WHAT DEGREE OF AGGREGATION?
The terms 'anonymisation' and 'pseudonymisation' describe the pro-
cesses now becoming widespread that are applied to health data, and
make it possible either to sever the link between the identity of the
person and the data about that person, or to be able to 'chain' the
information about that individual without knowing his or her identity.
Anonymisation (or de-identification) of personal data describes the
method and the result of processing personal data in order to irre-
versibly prevent the identification of the person concerned. In gener-
al, it is not enough to directly delete the elements that are themselves
identifiers in order to guarantee that any identification of the person is
no longer possible. An effective anonymisation solution must prevent
reidentification, which is not limited simply to preventing identifica-
tion (isolating an individual within a series of data, finding the name
and/or address of a person) but also correlation (linking together dis-
tinct series of data about a single individual) and inference (deducing
information about an individual from this series of data).
Pseudonymisation is a technique that involves replacing an attribute
(generally a unique attribute) by another in a record. The result of
pseudonymisation may be independent from the initial value (such
as in the case of a random number generated by the data controller
or a name chosen by the person concerned) or it may derive from
the original values of an attribute or a series of attributes, for example
using a hash function or an encryption system. The physical person is
therefore always potentially able to be identified indirectly. As a result,
pseudonymisation does not make it possible, by itself, to produce a
series of anonymous data; it reduces the risk that a series of data will
be correlated with the original identity of the person concerned. In this
regard, it is a valuable security measure, because it reduces the risks
for the person concerned, but it is not a method of anonymisation.
These definitions are in line with those stated in the Opinion of the

68
G29 Working Group of 10 April 201413
on the issue and of course with
those laid down in the new European Regulation on the protection of
personal data of 27 April 201614
.
With regard to the protection of personal data, pseudonymised data
remain personal data and are not intended to exclude any other data
protection measure.
It is interesting to note that in the text of the European Regulation,
which will soon become law in all Member States of the European Un-
ion in relation to the protection of personal data, pseudonymisation is
also covered in Article 6(4) on the lawfulness of processing as one of
the appropriate safeguards that the data controller may put in place
to justify the possibility of processing data initially collected for a given
purpose for another purpose considered as compatible.
This new notion of compatibility is important in legal terms to justi-
fy the further use of data for purposes different from those initially
envisaged.
The exponential availability of data from different sources in relation
to a single individual and the new capacities for processing these
data, and in particular Data Mining, considerably alter the concept of
reversibility of the anonymisation of personal data (reidentification).
Data considered to be anonymous when they are provided can
subsequently present a high risk of reidentification due to the
introduction of new techniques or new data sources, particularly in
a context marked by Big Data. The most secure anonymisation tech-
nique is still data aggregation, which transforms individual data into
collective data. But these techniques make numerous subsequent
processing operations impossible. It is therefore often legitimate to
retain the individual nature of the data while controlling the risk that
the persons concerned could be reidentified.
13. Opinion 05/2014 on Anonymisation Techniques.
14. According to the terms of Article 4(5) of the European Regulation, pseudonymisation
is 'the processing of personal data in such a manner that the personal data can no longer
be attributed to a specific data subject without the use of additional information, provided
that such additional information is kept separately and is subject to technical and organ-
isational measures to ensure that the personal data are not attributed to an identified or
identifiable natural person'.

69
Anonymisation (a reversible action) is desirable wherever possible
and provides impersonal data. In all other cases, the individual data
must be considered as pseudonymised (or indirectly identifiable) and
present a more or less high risk of both reidentification and disclosure.
It is the assessment of these two risks (reidentification and disclosure)
in relation to the sensitivity of the data processed that must guide the
implementation of appropriate security measures.
Is the distinction between pseudonymisation and anonymisation suffi-
cient in relation to Big Data?
Data Mining, which involves the processing of large-scale data using
powerful calculation techniques, uses either anonymised or pseu-
donymised data, and in very rare cases, if ever, directly identifiable
data.
For certain studies, anonymised data can have limited value in relation
to the intended objectives, and aggregation is insufficient to reflect
the health phenomena that only more detailed data make it possible
to reveal.
With regard to pseudonymisation, this technique can conflict with the
current conditions for compliance with the rules protecting personal
data as these are defined in France by the Law on Data Processing
Data Files and Personal Freedoms and soon in Europe by the above-
mentioned European Regulation.
How can we determine in advance the specific purpose for which we
want to collect and process data when the factor that specifically char-
acterises new research methods like Big Data is the processing of data
without knowing the purpose in advance?
In addition to the need to comply with this principle, there is an obli-
gation to complete the prior formalities imposed by the CNIL, which
are, as we know, complex and lengthy. The CNIL also has the option of
certifying the process associated with anonymisation of personal data,
for the purpose in particular of reusing public information placed
online15
.
15. Article 11 of Law No 78-17 of 6 January 1978, amended by Law No 2016-41 of 26

70
While the applicable legal texts do currently provide some options,
these remain limited and do not enable consideration of actual
requirements.
January 2016: 'It may certify or approve and publish reference guides or general meth-
odologies for the purposes of certifying compliance with this law in relation to personal
data anonymisation processes, in particular for the purposes of reuse of public information
placed online under the conditions provided for in Title II of Book III of the Code of
Administrative Procedure.'

71
ARTICLE VI
RELEASING DATA:
PATIENTS, USERS AT THE HEART
OF THE 'DISRUPTION'
Thanks to the collection and analysis of Big Data, techniques for pre-
vention, treatment, diagnosis and the monitoring of patients have
been changing at an accelerated pace, since mobile health especially
has been available to download on our smartphones.
The collection of these data, commonly referred to as Big Data, is most
certainly at the point of transforming the methodology used by re-
searchers and the approaches of the various institutional stakeholders
involved in managing health risk and expenditure in France.
While the notion of Big Data is now the 'fun' translation of the exist-
ence of intensive and industrial-scale health data collection, it does
not say anything about the procedures for accessing this information.
For this, another English phrase is often used: Open Data. This con-
cept, created out of a societal movement favouring maximum trans-
parency and civic democracy, is a response to users' 'right to know'.
Although that is not all it represents; Open Data in health is a philoso-
phy based on free access to digital health data, irrespective of the au-
thor. It assumes the structured dissemination of those data, according
to a method and an open licence guaranteeing free access and reuse
by everyone, without technical, legal or financial restriction.
Our law recently changed in order to modernise our health system
and redefine, on that basis, a process providing access to health data
held in the National Health Data System (SNDS), made up of the
following databases:
• the data produced as part of the Information Systems Medicalisation
Programme (PMSI), which reflects the activities of both private and
public healthcare institutions;
• the data from the National Interscheme Health Insurance Infor-
mation System (SNIIRAM) produced by the entities responsible for

72
managing a basic health insurance plan;
• the National Register of Causes of Death;
• the data produced by Departmental Centres for Disabled People
subject to the authority of the Independent Living Support Fund
(CNSA);
• a representative sample of reimbursement data by beneficiary pro-
vided by top-up health insurance entities and defined jointly with their
representatives.
The purpose of the SNDS is to provide data to contribute to:
• information about health and treatment options, medical and
welfare management and associated quality;
• the definition, implementation and assessment of health and social
welfare policies;
• an understanding of health expenditure, health insurance
expenditure and medical and welfare expenditure;
• information from health, medical and welfare professionals,
structures and institutions about their activities;
• health surveillance, monitoring and security;
• research, studies, evaluation and innovation in the areas of health,
medicine and welfare.
The data from the SNDS that are made available to the public are pro-
cessed into aggregated statistics or individual data created in such a
way that the direct or indirect identification of the persons concerned
is impossible. These data are made available at no cost. The reuse of
these data cannot have the object or effect of identifying the persons
concerned.
Access to the personal data held by the SNDS by research and analysis
institutions operating for profit (and in particular entities that produce
or market healthcare products, credit institutions, companies under-
taking direct insurance or reinsurance activities and insurance inter-
mediaries) may only be authorised to enable processing for the pur-
poses of research, analysis or evaluation contributing to one of the six
purposes listed above and corresponding to a public policy reason. In

73
this context, access is only authorised 'to the extent that these actions
are rendered strictly necessary by the purposes of research, analysis
or evaluation or by the missions entrusted to the body concerned',
and only the data necessary for that processing may be used. Beyond
that, a specific request for authorisation must be made to the CNIL.
The new National Institute of Health Data (INDS) receives requests for
authorisation, which must then be submitted for an opinion from the
expert committee for research, analysis and evaluation in the area of
health. This committee will announce a decision as to the methodol-
ogy chosen, the need for the use of personal data, the relevance of
those data in relation to the intended purpose and, where applicable,
the scientific quality of the project. At the initiative of the CNIL or at its
own initiative, an opinion may be requested from the INDS in relation
to the public interest nature of the research, analysis or evaluation jus-
tifying the request for processing. The CNIL will then assess the pro-
ject in terms of the principles relating to protection of personal data
and the value represented by the request.
The INDS will publish the authorisation from the CNIL, the declaration
of interest and the results and method.
The three main associations representing users (UNAF, FNATH and
CISS) have noted the need to include representatives of patients and
users on the strategic and monitoring committees to be created.
This is therefore how the French law envisages the operational di-
mension of Big Data in health, taking the view that research projects
implemented by profit-making entities must be subject to additional
constraints in order to access the data held by the SNDS because of
their commercial focus.

74
Press release from HDI, September 2016
ACCESSING HEALTH DATA IN FRANCE IS A LONG, COMPLEX AND UN-
EQUAL PROCESS
Paris, 26/09/16 - The Law of 26 January 2016 reforming our health sys-
tem demonstrates a commitment to opening access to health data, but
struggles to convince as to the objectives that France should be seeking
in this regard. The Healthcare Data Institute would like to draw attention
to the complexity of the mechanism introduced and has drawn up certain
proposals to resolve this situation.
The processing of health data is becoming a potential source of consid-
erable value creation, in particular for scientific research and the acquisi-
tion of new knowledge, from the management of care to the definition
of health policies By creating a new Title VI devoted to the provision of
health data, the Law of 26 January 2016 reforming our healthcare system
seems to be demonstrating serious ambitions.
'It should be noted that, in the current version of the text and before pub-
lication of its implementing decrees, the mechanism introduced does not
currently appear to be operational or able to allow the opening up of the
data that it has heralded, creating an onerous and complex procedure,
particularly for companies in the private sector', explains Isabelle Hilali,
Chair of the Healthcare Data Institute.
Increased complexity and absence of visibility
The members of the Healthcare Data Institute have observed an in-
creased level of complexity since the vote on the Law of 26 January 2016.
At least six steps must be overcome to obtain authorisation to implement
processing for research, and no less than six authorities may intervene in
the process (INDS, CNAMTS, CNIL, Expert Committee, research laborato-
ry or agency, trusted third party, etc.).
'The stakeholders, whether institutional or economic, need stability in or-
der to develop their missions and their business. It is difficult to lead a
industrial project or a major national project smoothly and with equanimity
in a legal context that is changing in reverse of other texts and, above
all, against the current tide of digital technologies', notes Jeanne Bossi

75
Malafosse, barrister, Counsel for DLA PIPER and member of the
Healthcare Data Institute.
Inequalities in access to data
For the members of the Healthcare Data Institute, inequalities in access
to data still exist in the new conditions introduced by Article 193 of the
Law of 26 January 2016. These demonstrate a lack of trust, or indeed
a mistrust, of private players, which are nonetheless at the very heart of
innovation in the healthcare sector, as has always been the case.
The potential risk is reflected in a decrease in France’s competitiveness in
terms of analysis of health data. Certain studies are currently being con-
ducted using foreign data, which are easier to access, and no infringe-
ments of rights and personal freedoms have been observed. There is also
a risk that parallel databases could develop, thus undermining national
ones, which have considerable scientific value.
Let us hope that the implementing texts for the law will introduce a gov-
ernance of the National Health Data System that is open, at a time when
the benefits come from the combined contribution of all private and
public stakeholders.
The above considerations have therefore led the members of the Health-
care Data Institute to request a review of these provisions.
Four proposals put forward by the Healthcare Data Institute
• Align the procedures between public and private stakeholders where
a public health interest is being pursued and where appropriate guaran-
tees are in place.
• Apply a process of control based on the principles that now drive
the protection of personal data in all European Union Member States
(Accountability and Privacy by Design): control a priori streamlined́ and
defined on the basis of an analysis of risk, control a posteriori reinforced,
much heavier penalties in the case of non-compliance. In this regard, the
possibilities for penalties imposed by the CNIL must be reinforced.

76
But will the SNDS really be the holy shrine of health data?
Thanks to the numerous smartphone sensors and the rampant spread
of health and well-being applications, health data are being dissem-
inated throughout a landscape that is much more vast and impres-
sionistic than just the SNDS, whose administrative design we have just
discussed.
The traditional proponents of Big Data, in particular the French
National Health Insurance Fund for Salaried Workers (CNAMTS) and
healthcare institutions, because of their coding, in fact no longer have
a monopoly on health data. This more widespread access has less to
do with Open Data than with the wide range of entities offering elec-
tronic services. Health data can take many different forms: medical
and administrative data in an original approach; raw and meaning-
ful data when they are prerequisites for apps that make it possible to
monitor physiological vital signs; to geolocate users in order to posi-
tion them in relation to a health institution; and to improve treatment
compliance, when this does not involve proposing a slimming coach,
a digital companion for better ageing, a connected night attendant,
etc.
One of the most pressing questions in relation to the data collected
in the context of the quantified self relates to the future of those data.
What will become of these masses of information collected with the
more or less informed consent of the users? Who will they actually
profit? And how can recourse to the concept of Open Data be valu-
able for these new uses, which we have been describing as such for
• Distinguish according to the type of data, in order to ensure that the
same constraints are not applied to data that have only a minimal risk
of reidentification. In this regard, we must not confuse the intensity of
the risk (the consequences of its occurrence) with the probability of the
occurrence of that risk.
• Ensure that the possibilities recognised by the texts for the CNIL to
simplify the procedures (single authorisations, reference methodologies,
etc.) are actually effective.
> Read the detailed position of the Healthcare Data Institute.

77
several years now?
The proportion of data provided by personal sensors out of all infor-
mation stored is expected to increase from 10% to nearly 90% over
the course of the next decade.
The challenges for the holders and hosts of all these data will be
measured on the basis of the profitability that these data represent for
the promoters of enhanced health and well-being, but not exclusively.
Because while the value of health data exceeds the value of bank-
ing data on the 'black market', this is because they have a particular
appeal for scrupulous and benevolent major accounts.
After the cyberattacks sustained by medical analysis laboratories or
health institutions, will the providers hosting the data required by
health apps be the next to be held to ransom?
Is the positive law robust enough to guarantee the security of the
'shattered' health data? More generally, are we equipped to take
advantage, collectively and individually, of Big Data while protecting
individual rights?
The new European Regulation on the protection of personal data,
published in the Official Journal of the European Union on 4 May
2016, which will enter into force in 2018, should make it possible for
Europe to adapt to the new realities of the digital world, by recognis-
ing the following in particular:
• The obligation to provide clear, intelligible, easily accessible infor-
mation for the persons concerned by data processing operations.
• The need for stated consent: users must be informed of how their
data will be used and must in principle provide their consent for
processing those data, or they must be able to object to processing.
The responsibility for proving consent falls to the data controller. The
realisation of this consent must not be ambiguous.
• The right to data portability making it possible for a person to re-
cover the data that he or she has provided in a form that can easily be
reused, and, where applicable, to then transfer it to a third party. This
involves giving people back the control of their data, and offsetting
some of the asymmetry between the data controller and the person
concerned.

78
• The right, for associations active in the area of protecting person-
al rights and freedoms in relation to data protection, to introduce
collective redress in respect of the protection of personal data.
• The right, for any person who has sustained material or non-material
injury as a result of an infringement of this Regulation, to obtain re-
dress for the loss sustained from the data controller or subcontractor.
This European Regulation is thus pushing forward the right of access
by individuals towards a right to portability by virtue of which they
now have a 'right of return' and a reappropriation of data relating to
them. Big Data is therefore becoming valuable for individuals, who
are becoming in part 'deciders' in relation to how their data are used.
Using this precious resource, there are numerous services that could
one day offer solutions with significant added value for individuals,
who have rightly been repositioned at the centre of these flows.
These new approaches, which are highly personalised, will reorien-
tate research towards solutions that are suited for individuals, their
lifestyles and their constraints, and will enable a sensitive and de-
tailed definition of their micro-objectives and motivational methods.
More collectively, data from Big Data, Open Data and Self Data could
prove essential for epidemiological research, because they will ena-
ble professionals to better understand the health status of the popula-
tion, to adjust the treatment administered to the patient by observing
models developed on a much broader scale, or to draw new conclu-
sions, for example in terms of the relationship between the course of
an illness and the associated environmental factors.
In addition, better use of health data could generate improvements
in productivity and cost reductions within the healthcare sector (in the
United States, the forecasts are for 300 billion dollars, in value, per
year).
There is still enormous uncertainty about how Big Data could change
our understanding of our relationship with health and how we can ra-
tionally and efficiently drive our health policy. But there is reason to
believe that we are standing on the brink of a fundamental revolution
that some people are referring to as a 'disruption', supported by the
digitisation of our service economy, by the prosperity of miniaturisa-
tion and by the substantial capitalisation of IT companies that bring in

79
their wake numerous innovative start-ups.
The rights of individuals must be seen as valuable resources in this
new economy that could be dominated tomorrow by the science of
algorithms. In any case, this is the basis on which Big Data will benefit
us all, and primarily the individuals themselves.

80
Created in 2014, the Healthcare Data Institute
is the first international Think Tank dedicated
to Big Data in the area of health, and acts as a
catalyst for ideas and projects focused around
Big Data within the health ecosystem.
The Board of Directors of the Healthcare Data
Institute includes representatives of Aviesan,
Axa, Caisse des Dépôts et Consignation, CEA,
CRI, Groupe Elsan, McKinsey & Company,
Orange Healthcare, Quintiles-IMS and Sanofi.
CONTACT
Pierre-Yves Arnoux
office@healthcaredatainstitute.com
+33 (0)6 07 13 77 13
@HCDATAINSTITUTE
healthcaredatainstitute.com
ABOUT THE
HEALTHCARE DATA INSTITUTE
HEALTHCARE
DATA INSTITUTE
21, rue Jasmin
75016 PARIS - FRANCE © RCA Factory 2016

White Paper HDI_big data and prevention_EN_Nov2016

More Related Content

What's hot (19)

Similar to White Paper HDI_big data and prevention_EN_Nov2016 (20)

White Paper HDI_big data and prevention_EN_Nov2016