Health information systems

From local to pervasive medical data

Christine Verdier
Lyon Research Center in Images and Intelligent Information Systems (LIRIS)
UMR 5205 CNRS, Claude Bernard University Lyon 1, Lumière Lyon 2 University
Ecole Centrale of Lyon
INSA of Lyon, Bat. Blaise Pascal, 7 av. Jean Capelle, F-69621 Villeurbanne cedex
[email protected]

ABSTRACT. This paper concerns the evolution of medical information systems. The evolution is
due to two main factors: the evolution of the society (a patient more concerned with his
health and more informed) and the evolution of Internet-based technology. We see a double
consequence: medical informatics must be pushed to medical professionals rather than the
opposite that prevailed until the nineties and health information systems (HIS) must move
from structuring specifications to visualisation and interface ones. The legacy health
information systems gather numerous data concerning patients with different structure, data
type, name, etc. To rebuild health information systems is not possible considering time-
consuming, conceptual and financial problems. But to rebuild a lifelong medical record is
also difficult and with no semantics guarantee. We propose in this paper to present what is
hoped about health information system design, what was really done and what we consider as
the future of this domain from legacy systems.
RÉSUMÉ. Nous proposons dans ce papier d’étudier l’évolution des systèmes d’information de
santé. Cette évolution est relative à deux facteurs : l’évolution de la société (le patient est de plus
en plus concerné par sa santé et plus informé) et l’évolution de la technologie, notamment
internet. Nous y voyons une double conséquence : l’informatique médicale doit être adaptée au
professionnel de santé contrairement à ce qui a prévalu jusque dans les années 90 : adapter le
professionnel de santé à l’informatique. La conception des systèmes d’information de santé doit
se modifier : de la structuration des données vers le développement d’interfaces de visualisation
et d’interrogation des données. Les systèmes d’information de santé existants concentrent un
nombre considérable de données de santé ayant des structures différentes, des types de données
différents, des noms différents, etc. Reconstruire les systèmes existants n’est pas possible au
regard du temps prohibitif, de problèmes conceptuels et financiers. Mais reconstruire l’histoire
santé d’un patient est également difficile et sémantiquement hasardeux. Nous présentons dans
cet article ce qui est souhaité en matière de conception de système d’information de santé, ce qui
a été réalisé et ce qu’il nous semble devoir être fait pour le futur à partir des systèmes existants.
KEYWORDS: medical information system, electronic medical record, adaptive interface,
pervasive data.
MOTS-CLÉS : système d’information médical, dossier patient informatisé, interface adaptative,
données pervasives.

SAS – 9/2006. Information décision patient, pages 87 to 108

88 SAS – 9/2006. Information décision patient

1. Introduction

At the beginning of the early research programs concerning medical information

systems (beginning of the 90th), the world of health was not really got used to
computer science applied to medical data. The early applications only concerned
management data, bed management, scheduling and so on. Two main problems can
be analyzed. First of all, the medical data are fickle, non-structured. Paper-based
medical records, medical summaries, notes, comments and other documents are not
easy media to be computerized. The second reason is that health professionals have
not considered, at least at the beginning of the computerization movement, it was a
priority. To moderate our language, we can say that the approach lead by the
computer scientist was not the better one: the difficulty to translate non-structured
medical document into structured ones, attributes with fixed forms and repetitive
access was certainly discouraging. Let us notice that, before XML, relational
databases were the only way to capture efficiently data.
First medical information systems were essentially based on the user interface
quality. This quality was interpreted as coloured, overloaded screens containing free
text fields to capture diseases, prescription. This type of interfaces is explained by
the difficulty to represent medicine through data model. The first designers were not
really competent in data modeling or not really felt concern about data
representation. They project onto the interface their own view of the medicine.
The second evolution was the prerogative of the computer scientist. The success
of the relational model had a considerable impact on medical information system
design. We, and many research teams, used the relational model to design the
medical record. Structured data and field were obtained to represent every item of
medicine. Prescription screens were built with a field for each item (posology, drug
name, drug indications, price, etc.). Diseases screens were described with the
problem-oriented medical record structure. These HIS were totally structured and
offer two real advantages: to make the data capture (pulldown menu for diseases
and drugs, check box, etc.) easy for the GP and to help for statistical studies and
epidemiology. The major disadvantage of these all-structured HIS were the use-
slowness because all fields have to be filled in.
The today representation of medical record is oriented to XML structure because
of its unstructured data representation. This language is well-adapted to medical
practice because doctors use many documents (operational report, clinical summary
of stay, drug prescription, etc.). A second advantage can be found: XML is an
interchange data format useful for communicating data through web applications.
We present in the first paragraph the definition of a health information system
and particularly what are the constraints for a good HIS. We continue with some
reasons for relative HIS failures. In paragraph 3, we will speak about related works
concerning HIS design and interoperability problems. We will go on with what we
consider as the future in medical computing: interfaces and grids.
 From local to pervasive medical data 89

2. Health information systems

2.1. Definitions

The field of information systems deals with systems for delivering information
and communications services in an organization and the activities and management
of the information systems function in planning, designing, developing,
implementing, and operating the systems and providing services. These systems
capture, store, process, and communicate data, information and knowledge. The
systems combine both technical components and human operators and users
(Davis). That definition enounced by Gordon Davis expresses the real environment
of information systems, composed of data types, technical aspects and wide-spread
of organisational effects. To make the users and operators efficiently communicate
through the information system (IS), three factors have to be considered: the data to
send, the time to convey, and the person to reach. Theses three factors can be
synthesize into a simple sentence: to give the good data to the good person at
the good time. This simple idea is opposite to the complexity of designing and
implementing an efficient (economic sense) information system that is able to
connect people (problems of systems heterogeneity), to deliver the data or
information necessary for decision support at the good time (not to late, not to
early). This triple constraint represents the foundation of a good information system.
It can be adapted to every domain of IS and is naturally true for health information
systems (HIS).
A HIS represents a class of IS. It includes the management IS that takes into
account the administrative part of the health domain and the medical information
system that covers medical, medico-economic and environmental aspects. That
second part will also deal with every connection data between the two categories
(Verdier et al., 1994). The medical information system (MIS) contains the part of
information which has a direct bearing on medical action and the automated
procedure for assisting users in decision making (Flory et al., 1997).

2.2. Description

An information system constitutes a coordinated whole of procedures, software,

materials and human resources making it possible to store, transform and share data.
Each element of this physical composition is useful to manage the organization as
well as possible to optimize the decision-making. The IS organization can be
represented by the following diagram:
90 SAS – 9/2006. Information décision patient

Managing system
Flows :
Data/information
Actors : Financial
Human Dataware Equipment
Software house Document
Hardware ERP
DB
Workflow/
groupware
Doc.
management

Process : I.S.
Data transfer
Security
Organisation
Activities system

Figure 1. Information system components

It does not seem useful to detail the components of IS of which description is in

all the good books in the field. Simply let us recall in some lines these components
in order to establish a parallel with the IS medical ones. (I)Actors: They are of three
different types; according to the terminology UML which widens the concept of
actor (compared to the Merise method) (Booch et al., 1998). Actors represent the
human beings, hardware and software which interact with the system.
(II)Procedures: they are diversified with the new complex aspects of IS; to the
traditional organisational procedures, we can add security settings (field very
significant since the introduction of Internet into the companies) and the procedures
of data transfer. (III)Flows: the systemic representation of the organization underlies
that it has a well-defined purpose and border. That also supposes that one can define
inter-systems and intra-systems interactions. The representation of flows thus
constitutes an important stage of the modeling of the information system.
(IV)Materials: The materials represent the physical part of an IS. They are of
different nature according to the user, the place of information collection, its
availability. We have little interest to describe the existing materials. Let us note
however that the awaited pervasivity of IS directs the material choices towards the
use of mobile tools (cellular telephone, portable computer, Wifi network, etc).

2.3. Conceptual specificities

Medical information systems were built gradually without very constraining

ministerial directives and official indications on medical information codification,
structuring of patient’s file or possibility of ambulatory medicine computerization. In
the absence of any regulation, medical information management was free for decision
makers, doctors and academics. If the result is overall interesting (innovative ideas,
 From local to pervasive medical data 91

adaptation to the users always closer to their daily job), the other side of the coin is
particularly heavy: systems developed are heterogeneous (as well at the conceptual
level as structural), awaited goals are different, numerous medical (or non medical)
partners with very dissimilar formation levels. That led to a plethora of information
systems (which besides are often reduced to a database) non communicating which
contain medical files entirely burst in the various places of care and which makes
absolutely impossible to reconstitute patient’s medical history. A vertical structuring of
the information system (by service) was privileged to a horizontal one (by patient).
Today developers of health information systems try to make up for time but the
rebuilding is difficult. A health information system contains several modules of which
the number and the importance are a function of the knowledge at the creation time,
the funds allocated and the partners implied in the IS.
– Administrative management of the patients: this part is always present and
preceded without any exception the design of medical information systems;
– Nurse file: in this module several evolutions were noted; from simple diary of
regulation to the control of the regulation;
– Patient file: this part never joins a design and content consensus;
The components of the MIS are similar to traditional IS. Only the trade
orientation changes the contents of the various components.
Actors
In medicine, the human actors are multiple. From this multiplicity is born
complexity. The wider the MIS is (MIS for ambulatory medicine vs MIS for
networks of care), the larger the number of actors is and their trades are varied. We
can count : doctors, paramedical professions, medical secretaries, non medical
actors of health: auxiliary of life, social workers, administrative staff, technicians.
The majority of these categories are listed in the CPS card (electronic card for
professionals of health). This card is used as entrance point with authentification
ofpeople in the MIS. Each speciality has a particular knowledge which is difficult to
share. It is thus very difficult to bind trades competences between them.
Hardware and software actors
We gather these two categories because they are closely dependent. The choice
of the materials is directed towards the mobile pervasivity and systems: wireless
technology is thus privileged and the mobile supports also: PDA, portable
computers, etc. The software actor most important is of course the medical file.
Consortia of standardization help with the formalization of the medical file by
provision of the community of the preset classes to improve reutilisability.
Procedures of security
They are essential and prone to controversy. At the beginning of the
computerization of health data, the principal handicap to which the scientists faced
was the security of medical files, protection against illicit intrusions and the risk to
92 SAS – 9/2006. Information décision patient

see the patient’s information medical seen and used by unauthorized people.
However, it is quite obvious that it was easier to seize a medical file paper on a desk
than to be introduced in an illegal way into an information processing system. It
does not remain about it less than procedures of security are fundamental in
medicine (as in any significant field). Several types of security are taken into
account:
– the European Commission recently authorize the encoding of health data in the
networks of care;
– the authentification with a coupling card CPS/Vitale card in France;
– the protection of networks with secure Intranet and firewall;
– the controlled access to medical data according to the degree of relevance and
emergency.
Several other systems start to appear but are used still little because in particular
of their dearness: biometric print, print of the iris.
Procedures of data transfer
These procedures will not be here detailed. They relate to the standardization of
formats and data exchanges. The reader can refer to the paragraph 3 which shows a
little part of data exchange for interoperability. Let us insist however on the fact that
these procedures are inherent with the width of IS and the will of its creators “to
open” their application.
Organisational procedures
The great difficulty for IS design will be today an organizational problem and no
more a technical problem. That appears quite as true to us in health. The principal
cause of countable deficit of health lies in the absence of optimization of material,
software and human resources organization. The relevant and efficient analysis of
organisational procedures will become a key factor of success in the future MIS.
Data flows
These flows are essential with the comprehension of the patient trajectory. They
make it possible to include/understand intra-organization and interorganisations
bonds. Only the modeling of flows intra-organization is succeeded. Within the
framework of networks of care, the representation of flows is much more difficult:
the process of care is nonlinear, the chronology brings only little diagnostic
elements and many feedbacks are to be taken into account.
Physical flows
The problems arise in a more crucial way within the framework of networks of
care where one externalise competences, care, administrative patient responsability
and the material used. To transfer a patient to his residence supposes to create a
hospital “out the walls” and thus to identify physical flows of drugs, disposable
materials, resources, etc.
 From local to pervasive medical data 93

Financial flows
The major problem of financial flows raises a major antagonism: the dichotomy
between the medical world and its financial support. The first systems separated the
MIS from the management IS in hospitals. Today, information systems include
medical and administrative data in order to be able to evaluate organisational,
financial and countable procedures.

2.4. Constraints for a good MIS

2.4.1. The first constraint: a good information

The building of a real good HIS is totally inseparable from capturing data
directly from the source. It means that other copies of the same data lead to
mistakes and errors. The second way for good information is to imply health
professionals with continuous feed-back, daily help in their job. A necessary
adaptability of the computerized system to the health practitioner (and not the
opposite) must also be done to simplify the interface and the communications
between the machines and the users. On-line help system has to be added to the
application. This help must be easily understood to avoid mistakes and
consequently, annoyance, discouragement and then rejection of the system. Let us
notice that the impact of interface quality is not so sizeable according to European
countries. In France, it affects directly the use of medical computerized systems. To
be really adopted (of course without any legal constraints), a computerized system
must be very close to the natural language, the mental behaviour of the user and his
meaning. Today systems come close to this constraint but are note totally human-
aware. We propose some ideas in paragraph 4.1.
The good information constraint consists also in the quality of structuration.
We will speak in depth about the structuring versus the non-structuring of medical
data in a further paragraph. What is the real problem? To link the freedom of
manual data capture and the necessity of structured data for automated treatment.
The former ideas focused on the 'all-structured' to help for epidemiological studies,
statistic reports, SQL queries and all types of computerized processing. We think
that this way is really and definitely the only one to support quantitative queries. But
the antagonism is then obvious. The real job of a medical professional is to care,
nurse, relieve, help but surely not to capture every piece of the medical information
he produces in formatted electronic documents. So, either the system automatically
translates from paper-like document to receive structured data without a manual
intervention, or the partition of the medical information must be so easy to
understand that any real mental effort is needed. In any case, the structuring must be
a final goal even if its not the former way for data storage.
Semantics of data is also a guarantee of quality. The huge of medical
information systems, electronic records, Web-based documents pose two main
problems: data transfer -how preserve data semantics during the transfer from a
94 SAS – 9/2006. Information décision patient

system to another?- and data representation -how to create the data model before the
programming step? According to us, semantics represents the main point of HIS.
The degree of confidence in the data obtained (to care) directly from a database or
from several data sources, raw data or treated data must be 100%. An information
must be always exact, exhaustive, complete and confident. The query result: 'the
patient Jack has a diabetes' is true to 95% can't be accepted. The medicine is not an
exact science but to increase the uncertainty is inconceivable. This reality can
partially explain the rejection of some medical information systems (particularly in
France). So semantics maintenance of medical data must be a real priority in the
design of information systems.
To plagiarize many authors, we can say that: too much information kills
information. If semantics is essential to correctly understand medical information,
the volume of this information must be studied. Relational databases limited the risk
to loose the user among a huge a data. The important rising in the use of the Web
increases the risk considerably. Query systems must be improve to fit this new
challenge. What does it mean? 1) The definition of new query systems (SQL-like
systems are only really efficient for structured data) even if some SQL-modified
languages try to settle some problems linked to Web data retrieval; 2) The use of
data semantics and domain ontologies to filter data; 3) the definition of users
profiles and 4) the taking into account of time.

2.4.2. The second constraint: the good person

This aspect of quality consists of two main objectives: to find the good medical
partner and to verify his ability to taking care of the patient.
The first aspect deals with the identification of the health professional.
Traditionally, in hospital or at home, a patient is referenced by a doctor. But to
maintain the continuity of medical action, it is really necessary to track the patient
care. So to connect several health professionals between them can be helpful for the
patient care. This problem is really acute when we consider home-hospital networks
and particularly home care. This type of care will be developped in the further years
because of economic reasons (it should be less expensive to care people at home
rather than in hospital) and psychological reasons (it’s often better to be at home for
a patient rather than in hospital). The home care supposes that many persons are
close to the patient for caring, helping, advising and globally participating to the
patient health trajectory. To couple different jobs to better identify the health
process and produce a better care following, these different jobs must be linked and
each person must have a small knowledge of the other jobs. We can divide the jobs
in the care process of two types : the medical professionnals and the non medical
persons (social workers, household help, auxiliary of life). The recovering of
knowledge is important to allow the care following. So to identify the job and the
knowledge level of every person in the care process is essential to connect different
medical actions. To identify the good person with the good information, it is
necessary to adapt the today human interface to include non professionnals.
 From local to pervasive medical data 95

The second aspect concerns the ability to access medical information. Thus,
connected to the nature of information, access rights must be taken into account
(with biometric measures or login and password). This part of the security acess is
difficult because of the multi-partners of the IS and the huge data of the medical
record.

2.4.3. The third constraint: the good time

To receive information too late or too early is similar and negative. The time
must be considered with a multi-faceted view.
What must be considered first is the time to distribute information. This time is
often difficult to analyze because the care to patient is health state-linked. A patient
having a suspicion of myocardial infarction must be dealt with in less than one hour.
It supposes that information absolutely needed is immediately available (allergies,
essential medical past history, etc.). An information system must be useful for that
and adapted to emergency. Let us notice that this situation is possible when the
patient is 'captive'. It means that the patient is knew in the hospital, his medical
record is up-to-date and querying the system give an immediate good answer. For
nomad people, the situation is not the same, more crucial and more difficult.
Pervasive information systems should give a better answer to that particular
problem (particularly with data grids). But let us consider the current problem. A
today solution can be found in health cards. In France, the Vitale card exists and
contains the administrative part of the social security information. A further version
should integrate some essential medical information. We must add that some
particular health cards already exist for particular chronic disease or health
symptoms (cards for penicillin allergy for example). The nature of medical
information, we just felt that, is conditioned to time. Emergency information must
be obtained as soon as possible. Treatment or information of monitoring could be
researched with a more long time and synthetic information is not time-acute. So,
how can we reconcile the time-adaptability with a patient health fluctuation?
The second view concerns the representation of time in information systems. If
static models are well-known and -used, dynamic aspects stay hard to model. UML
proposes a whole of models tools to describe the organization (Booch et al., 1998).
Some models like behaviour diagrams are useful for dynamic representation but
most difficult to implement especially in medic area. Let us explain in a few words.
The medical time is really difficult to represent because of its great modularity.
Some papers have been written about medical time (for serializing data), about time-
based database but not really, according to us, to the design of time in medical
information systems. Why? Firstly because of the variability of the time-notion. The
time-measure of the blood pressure depends on the patient state: a non-pathologic
patient (without hypertension) can have a blood pressure measure twice a year. At
the opposite, a hypertensive patient measure must be done once all 10 minutes. To
represent the time with a month/year/day format is insufficient. Consequently, the
comparison between time-data series is not easy. Approaches concerning time-
96 SAS – 9/2006. Information décision patient

scales, temporal databases are developed and give some interesting results with
Allen algorithm and closely related algorithms.
The third aspect concerns the delay to take in charge a patient. This time is
linked to the first aspect, the time for delivering good information. This facet is
highly correlated to at-a-distance medical care and telemedicine. Information
technology, networks, and all telematics-based applications open a new challenge in
patient care. Current medical information systems are used for inpatients. The
enlargement of places of care, the need for at-home care, for any-place-of care, lead
to think about another approach of information systems with so-called
communicating information systems. New tools help in creating at-a-distance
information systems because new technologies exist: wireless networks, satellite
transmission, medium waves transmission, shared medical record, dynamic and
static design models, and so on. Main problems are the cost of the modification of
current IS, politics' will and finally a real large consultation to connect people.

2.5. Reasons for relative failures

All job-wide firms have IS to manage data, to improve strategy, to make

decision easier and better. The use of IS is based on consensus and could not be
discussed. The attempts of IS in health have until now failed. If some parts found
solutions, the whole is not powerful. The reasons are plethoric and of different
nature. Most important seem to be the great difficulty of representing a complex
world in perpetual evolution. Complexity can be apprehended on several levels:
– The fickleness of medical information: the data “loss of visual faculties” does
not produce same information according to whether the patient is diabetic or not;
– The absence of standardized vocabulary: except the International
Classification of Diseases( ICD 10 – WHO), no agrement arrived on job-oriented
classifications;
– The difficulty of representing the semantics of medical information through
multicriteria research;
– The absence of consensus on the structure of medical record: opposition
between problem-oriented structuration and chronological structure;
– The difficulty for removing the no more relevant information from the medical
record: the relevance of medical information decrease with the time, except for
specific medical history;
– The search for information useful for the medical follow-up of the patient:
made difficult by the bursting of the medical record between various places of care;
– The representation of various jobs in which each one can cover different
specificities: liberal nurse and nurse of surgical units;
– The difficulty of defining IS borders and its universe of discourse.
 From local to pervasive medical data 97

These some factors, non exhaustive, show at which point this field is delicate to
represent. Data processing likes structured data and little changing but health
represents the opposite.
Let us take again the above mentioned factors. Each point concerns a particular
sector of scientific research in data processing and it is undoubtedly the principal
consequence of the quasi-failure of IS which always considered only one facet. The
factor (1) concerns management and the capitalization of knowledge, factors (2) and
(3) relate to ontologies and representation of data, points (4) and (7) can be solved
thanks to modeling tools, factor (5) is a basic problem of data and point (6) relates
to information search, adaptive interfaces and “intelligent” data query. So how to
imagine the construction of a medical information system which does not take into
account all these aspects ? The major difficulty thus consists in federating these
elements to offer an information system oriented-trade where each partner will be
able to find information he is looking for whereas it is located in the larg Web
environment. The concept of patient care process is emerging which is called
patient’s trajectory. Let us say that some authors see a difference between care
process and patient’s trajectory. We use the both terms because it represents for us
the interval between the patient enters the health system and the date he leaves. So
medical information systems must imperatively provide a relevant visualization of
the care process.
Let us turn some pages of history. In first applications of systemic to IS, the
functional company division prevailed because it was representative of the idea that
one was done of this world cutting in engineering and administrative departments
(accounting department, store, direction departments) or/and in functions (financial
or accountancy function, purchase function, etc.). This Taylorian cutting made it
possible to identify the borders of each company’s subsystem and thus to apply the
traditional profitability and evaluation indicators. The more complex organization of
current companies, their geographical, thematic, functional and job bursting change
completely the organization of IS. The 2000s’ years mark an evolution of
companies’IS perhaps as important as the technological revolution born after the
second world war. The health domain copies the same characteristics and the same
evolution. If in hospitals, always prevailed a cutting in services (for political or
medico-economic justifications) and event if this phenomenon must go on, the way
of apprehending the MIS must be deeply anchored on the concept of care process
i.e. transversality (or patient’s trajectory).
Another reflexion must be carried out which also impacts the functional
architecture of traditional IS. We have traditionnaly copied medical information on
the human body. Classifications of diseases, medical records were created according
to the organic cutting of the man: heart, lung, kidney, etc. This categorization of
medical information is explained easily by the medical training: each medical
discipline is specialized in a body part (brain for psychiatry and neurology, heart for
cardiology, kidney for urology and nephrology, etc.) and the covering of knowledge
between different medical specialities is very limited. The disadvantage in this way
98 SAS – 9/2006. Information décision patient

of organizing medicine is however counterbalanced by the facility of the training

and the knowledge of the field. Organic cutting was thus naturally reflected at the
software level. Computerized medical records, heart of the medical IS, were organ-
oriented (cf figure 2).

Cardiology medical

Traumatology medical
Pneumology medical

Nephrology medical

Figure 2. The patient = a whole rather than a sum of body’s parts

We obtain inevitably intersections between this various softwares because of the

necesseraly duplication of information. This representation raises three essential
questions:
– How to identify the patient as a whole whereas he is represented only with a
sum of body parts;
– How to identify the care process since we add to the geographical dispersion
of the care, a anatomical distribution of medical information?
– How to replace the patient in the middle of the medical IS “whereas he enters
the health system part of body by part of body”?
The patient represents an entity and not a sum of body parts. We see it well, it is
no more possible today, considering the wideness (extend) of the medical
knowledge, the number of medical treatments and the increase of the population
targeted to consider this functional design.

3. Related works

The structuring of the medical file is probably one of the most difficult
exercises. The major difficulty is due to several phenomena. Medical information is
complex, changeable according to various criteria (place, time, for example), related
to a given trade (an urologist does not use same information as a general
practitioner), retroactive and evolutionary. We can add to this list the absence of
precise definition of the medical file between ambulatory medicine and hospital
medicine or within the same structure between various doctors. Contents of the
medical file are specific to each doctor. The absence of real obligation of sharing
medical file, directives and other constraints involved a paer-based solution which
 From local to pervasive medical data 99

was at the same time perfect for the doctors and completely unusable for a data-
processing traanslation (the textual support does not make it possible to make
research on the data). The impact of health networks made become aware the
importance to share data, so to structure and model the medical file. Teams worked
on these problems. We can quote works of J Cimino (Cimino, 1996; Barrows, 1994;
Cimino, 1994) on the modeling of classifications and their inclusion in the medical
file. We can also quote works of J.R. Scherrer with the systems Diogene and Galen
(Borst, 1995; Alpay, 1995), (Scherrer, 1995; Bréant, 2000) in Geneva hospital.
French teams were also interested in this work: P. Degoulet’s team in Paris and
M. Fieschi’s team in Marseille. These two teams regularly collaborated: (Soula,
1997; Burgun, 1995; Joubert, 1994; Degoulet, 1997; Engelmann, 1995; Lavril
1994).
The second part of the research works about medical information systems has
concerned interoperability. Many computer-based medical applications were created
for particular needs in closed environments. The first consequence and undoubtedly
most serious was the total absence of communication between heterogeneous
systems. In a completely antagonistic way, the need for creating shared medical files
made essential the cooperation between heterogeneous information systems.
Works concerning the structuring of the medical file and interoperability
between the medical systems became very important. Great projects were started to
work in this field and to try to find a structure adequate and consensual medical file
and related applications. We arrived at the middle of the Nineties with a spectacular
profusion of research projects in this field. Among these projects, we can notice the
European project AIM (Advanced Informatics in Medicine) which proceeded
between 1991 and 1995 and take the first steps towards the computerization of
medical files and to propose European and viable solutions. Several projects started
from AIM consortium.
The goal of HELIOS project (Lavril, 1994) was to conceive an environment of
development intended to computerize the medical applications. The principal idea of
the project was to offer a set of tools ensuring interoperability between the
applications. The awaited goals were: (I)to create an opened and modular
environment of development to facilitate the production of medical applications,
(II)to use a unified mechanism to integrate the medical application software; (III)to
primarily target the environment towards the automation of the operating room;
(IV)to support the reutilisability of medical applications. HELIOS was conceived in
oriented-object.
The goal of project GEHR (Good European Health Record) was to define an
architecture standardized for the computerized medical files (electronic medical
records). It included storage, communication of medical information between
doctors, hospitals and analysis laboratories. The discussion thread of the project was
to ensure a perfect portability in order to abstract itself from problems of coding,
language or operating system. The result of project GEHR was concretized by a
100 SAS – 9/2006. Information décision patient

multi-media architecture of data for the use and the sharing of medical files by
taking into account clinical, technical and medico-legal functionalities.
The goal of the Menelas project was to carry out a system of access to medical
files based on natural language. The idea of this project came from the observation
of medical behaviors. Medical data are seized in the form of free text in medical or
operational summaries. The Menelas project thus had two essential goals: to provide
an access simplified to medical files in free text and to improve the multilingual
access to medical nomenclatures.
Other projects were born at the international level: we can quote project W3C-
EMRS and project HELP.
Boston’s hospital has created a common medical file (W3C-EMRS) which
specifies the types of data patient having to be transferred in the common file to
ensure the continuity of care. The principal idea was to ensure a longitudinal sight
of the patient care and to ensure a coordination of the medical actions. To achieve
this goal, the authors translated the specifications into HL7 messages.
HELP system (Kuperman, 1991) is an information system based on knowledge.
It included all the elements of taking into account and follow-up of the patient and
integrated an engine of decision-making aid. The decision-making aid was done in
an interactive and asynchronous way starting from data and from the management
of effective time. HELP system was based on a basic architecture of data which was
used also for the structuring of patient files.

4. Medical information system for the future

Many research tasks are to be carried out to thwart the negative impact of
heterogeneous systems. A possible solution which is the subject of many
publications consists in building platforms of data exchanges which are useful of
“translator” between the various information systems. Other solutions, which we
privilege, consist in using existing systems and moving the problem of
interoperability on the level of the interface. It is thus a question of creating
interfaces able to go to seek medical information in the data sources and to rebuild
on the doctor computer the medical file which he needs to work.

4.1. Iinterfaces adaptation to users

4.1.1. Generality
The heterogeneity of medical information systems shows a double antagonist
consequence: the impossibility of changing well-used legacy systems because of
cost-, and time reasons; and the necessity of sharing medical file to carry out the
patient care. Many difficulties can be noticed: heterogeneity of the data structure
 From local to pervasive medical data 101

(XML, RDB, OODB), heterogeneity of the content (num_pat vs n°Patient),

duplication of data, lack of standardized vocabulary (ICD, ICPC, others thesauri).
These difficulties must find different solutions according to each problem:
ontologies to thwart heterogeneous vocabularies, specific algorithms to detect a
given patient (with his name, first name and adress), mediator to translate
heterogeneous data sources in a pivot language, etc.
If we cannot change MIS because of previous reasons, we can use the different
technologies to overcome the problem and to solve it at the interface level. Doctors
are not computer scientists and to move the heterogeneity problem from the DB
level to the interface does not mask the problem. So an interface-based approach has
to mask the difficulty to the user and propose user-friendly and ergonomic graphical
user interfaces. Three concepts have to be considered: the presentation of the
interface as a web-based application with easy navigation through web pages, the
adaptation to the user profile (according to access rights –very important in
medicine-, medical specialty, etc.) the real-time adaptability to the user browsing
(MIS learning, increasing of the knowledge, etc.).
We said that to move the heterogeneity problem to the interface does not mask
the difficulty to retrieve, remove and share medical data: it is a crucial problem.
Mediators and exchange platforms propose solutions to translate relational, object-
oriented and XML data into a pivot language or into XML language. Solutions exist
and are correct because it is less difficult to transform very-structured data (like DB)
into semi-strutured or unstructured data than the opposite. The trickiest problem is
the semantic one. To rebuild the medical record of a patient whose cares took place
in different places of care is really difficult because the identification of the patient
is not obligatory or different from a place to another, so to be sure to link medical
data of a unique patient and not to mix data from different patients is very difficult.
Many research projects work on this problem. We can cite for example, what is
done in (Quantin, 2004; Kelman, 2002). Numerous works are led about semantic
web concerning particularly ontologies, interoperability, facetted databases and
context-based semantic. Among these works, we can cite (Widhalm, 2003; Laks,
2003), (Ross, 2004) and (Zhu, 2004).
We proposed a new solution based on a coupled Topic Maps and Description
Logics approach to overcome the semantic data integration and a graphical interface
based on cartography to simulate the web navigation.

4.1.2. Concepts cartography

General remarks
The reconstitution of the medical history of a patient becomes then a challenge
except to build ad hoc systems where each element is verified again (at the
conceptual and semantic level). For these same reasons, distributing medical
information from heterogeneous sources and leading epidemiologic studies
according to stratification criteria is not possible without manual intervention.
102 SAS – 9/2006. Information décision patient

These two problems led us to think of another mode of interrogation. Since the
ad hoc rebuilding is impossible (except to carry out expensive handling), it seems
preferable to build a system downstream with a complete freedom for the user to
question these data without being concerned with constraints enumerated above.
For that, we built a system of interrogation/navigation based on Topic Maps.
Data structure and values are represented in a single format representing navigation
cartography. This double system makes it possible to handle medical information
according to two different manners: adapted to the user profile and adaptive in real
time with navigation. It produces two results:
– A centred-patient navigation allowing rebuilding his medical history;
– A centred-population navigation allowing incorporating statistical data for
epidemiological studies.
Topic Maps approach
Topic Maps represent a formalism of management, representation and
organization of knowledge (Sigel, 2000). It is used to formalize human knowledge
with an aim of facilitating and optimizing the operations of creation, management
and search for this knowledge. Topic Maps represent knowledge in a related graph
composed of nodes bound by semantic relations. Topic Maps thus make it possible
to represent knowledge but also to organize resources. The basic concepts of Topic
Maps are very simple (ISO, 1999). Topic Map is represented by a graph of topics,
which constitutes subjects, linked by semantic associations. Topics associated
resources called occurrences. To exchange Topic Maps, TopicMaps.Org Authoring
Group standardized in 2001 a XML-based grammar for serializing Topic Maps.
This grammar is called XTM (XML for Topic Maps).
Thus, Topic Maps constitute a very powerful formalism of representation. It
makes it possible to represent relational databases, object-oriented databases or
XML files with the same flexibility. The great force of the formalism led to its
antagonism: it is not possible to represent hierarchies between the topics (in a
universe where all is topic). Topic Maps are coupled with description logics which
makes possible to obtain, at the same time, flexibility and simplicity of
representation (Calvanese, 1999; Baader, 1999).
This information system, intended for the end-user, provides an interface of
navigation/interrogation of multi-source data enabling him to free itself from the
localization of data, of their structure and their nature. A session of
interrogation/navigation is concretized by the elaboration of a document non
persistent and formatted according to users’ preferences. This document is called
virtual document because (especially for legal reasons), it is not stored in the
system.
Centred-subject navigation is applied to the interrogation of a subject. The
subject can be anything: Joyce, blood analysis of Robert or the Boston hospital.
From the subject chosen, the navigation/interrogation continues stages in stages and
 From local to pervasive medical data 103

every link between two subjects contains the cardinality of the set: we speak about
exactly instantiated graph. The result of a browsing is represented by a personalized
and virtual document.

Figure 3. Cartography centred-subject

The second mode of navigation consists of dividing the whole of data sources in
order to carry out statistical groupings. This interface is not used for navigation in a
space of data. The user visualizes data analysis in a hierarchy of concepts which
expresses semantic relations and data distributions. From the Topic Map, the user
explores the concepts and specifies the analysis attributes. The system calculates the
distribution and returns the results. This functionality leads to data partition
according to a particular interest and with immediate result displayed on the screen.

4.1.3. User profiles

To improve our idea, we work on the concept of user profile in order to give
fluidity to the inter-job communication. The first user-profile application is to
remove not-relevant information according to the context (defined by the scope
concept in Topic Maps). Indeed, all users of the MIS do not have the same waitings,
knowledge, competences, centers of interest, etc... So it is necessary to define
criteria or semantic rules for each user or user group. Two essential points must be
104 SAS – 9/2006. Information décision patient

taken into account: the construction of user profile and evolution and update of this
profile. With the idea of interoperability, profile contents to be built will be
characterized by a whole of open dimensions. Until now, we distinguished seven
dimensions:
a) Professional data: they include information about identity, profession, main
activity, place of work of the user. These data are relatively stable.
b) Preferences: this dimension represents specific preferences i.e., centres of
interest (sport medicine, dietetics, etc).
c) Level of knowledge: this dimension provides information about the
knowledge level of users. Two users who have the same activity do not have
necessarily the same competence level.
d) Domain ontology: to allow the translation of a medical term in another
simpler for a user with fewer competences (e.g., Myocardial Infarction
Heart
Disease).
e) Authentication and confidentiality: this double dimension represents the
access right to medical information (for each user and place of care) and data
protection (cryptography).
f) Preferences feedback: the behaviour of the user is identified during the
navigation or the interrogation of the system.
g) Miscellaneous information: this dimension represents information that can’t
be classified in other dimensions.

4.2. Medical data grids

4.2.1. Definition
Pervasive information systems are difficult to define. It probably concerns several
information systems (not only once) and the definition is different according to the
point of view which one adopts. If we try, we can say that a pervasive information
system is a collection of communicating information systems that are able to produce a
data for a good person at a good time in a good place on a wished computing device.
According to E.J. Sol (Sol, 2001), people will use within 10 years, hundred
interconnected computing devices around them, a majority of wireless connections and
many embedded devices. It supposes that major data-processing problems in the years
to come are: the security, the replication and the resilience. The security problem is
double: to guarantee the sense and the semantic of the data during the transaction and
to ensure that the data is only viewed by the authorized person.
A direct consequence of pervasive information systems is pervasive computing.
In (Cherniack, 2001), the authors consider that pervasive computing make a
computer so imbedded, so fitting, so natural that we use it without even thinking
about. They consider that pervasive computing has to force the computer to go to
people and not the opposite. These definitions meet on the crucial point: to bring
 From local to pervasive medical data 105

computerized data at any moment at any place on any device to the person who
needs it. Medical domain undergoes the same evolution. Some articles are dedicated
to this evolution of health information systems. So, it is obvious that contrary to
other domains where computer science has gradually evolved, health domain has a
great jump to do to meet this challenge.
The main way to adapt pervasive computing to health area is the health care
networks. A health care network can be defined as a link between health
professionals for a disease. This link represents the whole data useful for diagnose,
understand, care and follow a particular disease. Many health networks are created
in France and in Europe to manage complementary health care around a particular
disease: in France, are proposed health networks in kidney failure, pregnancy,
measles and so on. The networks group together different health professionals and
are realized with particular computerized systems. Many problems can be raised:
every new networks develop its own structure but use similar conceptual entities, to
make heterogeneous information systems is very difficult, and to be sure to link
semantic-garanteed data is practically impossible to do. A solution can be found in
data grids. More and more works deal with data grids. Main works concern image
processing and retrieval on grids (Montagnat et al., 2005).

4.2.2. Data access and patient identification

Data access represents a specific problem in grids. The data are located in
distributed information systems with no idea of place of storage. Some metadata are
stored in the local cache memory but to find a data set coming from different data
sources is difficult. In medicine the problem is more crucial because the
identification of a patient is non existing or different from an IS to the other. So to
rebuild the medical history of a patient without a correct identification number is
very difficult. We try to adapt existing algorithm (Quantin, 2004) to grid
problematic. The architecture of our system is the following:

Q1 : patient search by name, first name,

date of birth and address

R1 : research in the local cache, connexion to

Mediator
the DSn and retrieving of the data

LC LC LC LC
Application of data comparison algorithm
and result

Ds1 Ds2 Ds3 Ds4

DSn : data sources

LC : local cache
Q : query
R : reply
: persistent link
: dynamic link
106 SAS – 9/2006. Information décision patient

Figure 4. Data retrieval based on grid architecture

Query Q1 is sent to the mediator which role is to translate the query into a pivot
language to be understood by local cache. Every local cache containing metadata
corresponding to the query serach for the data in the local sources. After returning
the reply, many data can be concerned with the query. Then, we apply a data
comparison algorithm to compare all the character strings and give the good patient
according choosen criteria. The reply is translated by the mediator in a pivot
language to be used subsequently for any need (transmission of data, statitiscal
need, etc.). All queries (Q1...Qn) are sent in a dynamic view to have a persistant and
relevant result. This dynamic part is not discussed in the paper.

5. Conclusion

We have presented in this paper some discussion about the future of medical
information systems. What we think about this is mainly the ability of new systems
to deliver the good information to the good person at the good time on the good
device.At one side, MIS can’t be anymore static, totally structured, non flexible to
each user, uniform and non changing. At the other side, MIS can’t be easily
modified to take into account new challenges. A solution is to tranfer the problem
from the conceptual level to the interface level. At this level, two possibilities
coexist: data access from heterogeneous information systems through data grids and
adaptative interfaces to navigate in a data space and query data with a complete
semantic guarantee. We have presented in few ideas basic research ideas of our
team work. The reader can find more precisions in the following papers (Ouziri,
2005), (Ouziri, 2003).

6. Bibliography

Allen J.F., “Maintaining knowledge, about intervals”, Communication of the ACM, 1983,
(26), 11, pp. 832-843.
Alpay L.L., Lovis C., Nowlan W.A., Solomon W.D., Rush T.W., Baud R.H., Juge C.,
Scherrer J.R., “Constructing clinical application: the Galen approach”, Medinfo 1995 ;
8 Pt1 : 65-9.
Baader F., “Logic-based knowledge representation”, M.J. Woldridge and M. Velose eds,
Artificial Intelligence today, Recent trends and developments, 1999, pp. 13-41.
Barrows R.C. Jr, Cimino J.J., Clayton P.D., “Mapping clinically useful terminology to a
controlled medical vocabulary”, Proc Annu Symp Comput Appl Med Care, 1994, 211-5.
Booch J., Rumbaugh J., Jacobson I., The unified modeling language : user guide, Booch,
Jacobson, Rumbaugh eds, Addison Wesley, 1998.
Borst F., Griesser P., Bourdilloud R., Scherrer J.R., “Fifteen years of medical encoding in the
Diogene HIS”, Medinfo, 1995 ; 8 Pt1 , 43-6.
 From local to pervasive medical data 107

Bréant C., Borst F., Campi D., Griesser V., Son Le H., Junod J.M., « Expanding Diogene
with a clinical information system based on a new hospital-wide clinical findings
dictionary”, Int. Journal of Medical Informatics, 58-59, 2000, pp. 167-177.
Burgun A., Botti G., Bodenreider O., Delamarre D., Leveque J.M., Lukacs B., Mayeux D.,
Brémond M., Kohler F., Fieschi M., “Knowledge acquisition from the UMLS sources :
application to the description of surgical procedures”, Medinfo, 1995 ; 8 Pt 1 :75-9.
Calvanese D., De Giacomo G., Lenzerini M., « Representing and reasoning on XML
documents : a description logic approach », Journal of Logic and Computation, 1999,
Vol. 9, No. 3, pp. 295-318.
Cherniack M., Franklin M.J., Zdonik S.B., “Data Management for Pervasive Computing”,
Tutorial, proceedings of the 27th International Conference on Very Large Data Bases,
Sept. 11-14, 2001, Roma, Italy, available on
http://www.vldb.org/dblp/db/conf/vldb/vldb2001.html
Cimino J.J., “Review paper: coding system in health care”, Methods Inf Med, 1996, Dec, 35,
(4-5), pp. 273-84.
Cimino J.J., Clayton P.D., Hripcsak G., Johnson J.B., “Knowledge-based approaches to the
maintenance of a large controlled medical terminology”, J Am Med Inform Assoc, 1994,
Jan-feb, 1, (1), pp. 35-50.
Davis G.B., “Information systems conceptual foundations: looking backward and forward”,
chapter 5, available at http://www.terry.uga.edu/~ekarah/davis.pdf
Degoulet P., Fieschi M., “Critical dimensions in medical informatics”, Int J Med Inf, 44, (1),
1997 Mar, pp. 21-6.
Engelmann U., Jean F.C., Meinzer H.P., Schröter A., Günnel U., Demiris A.M., Schäfer M.,
Evers H., Degoulet P., “Integrated image processing in clinical applications : the Helios
approach”, Medinfo, 1995 ; 8 Pt 1, 411-5.
Flory A., Verdier C., Mercatello A., “Conceptual modelling of medical information”, Health
and System Science, Hermès eds, Vol. 1, No. 3-4, Dec. 1997, pp. 263-90.
ISO/IEC 113250, Topic Maps, Dec. ISO/IEC FCD, 1999.
Joubert M., Fieschi M., Robert J.J., Tafazzoli A., “Users conceptual views on medical
information databases”, Int J Biomed comput, 1994 Oct, 37, (2), pp. 93-104.
Kelman CW, Bass J, Holman CDJ., “Research use of linked health data - a ‘best practice’
protocol”, Australian and New Zealand Journal of Public Health, 2002, 26, (3), pp. 251-
255.
Lakshmanan L.V.S., Sadri F., “Interoperability on XML Data”, Lecture Notes in Computer
Science, Publisher: Springer-Verlag GmbH, ISSN: 0302-9743, Vol. 2870/2003, pp. 146-
163.
Lavril M., Dore L., Zapletal E., Jean F.C., Degoulet P., “A reuse oriented development
database : the Helios object information system”, Comput Methods Programs Biomed,
1994 Dec; 45 Suppl, S35-45.
108 SAS – 9/2006. Information décision patient

Montagnat J., Bellet F., Benoit-Cattin H., Breton V., Brunie L., Duque H., Legre Y., Magnin
I.E., Miguet S., Pierson J.M., Seitz L., Tweed T., “Medical images simulation, storage
and procesing on the European Datagrid testbed”, Journal of grid computing, Kluwer,
Vol.2, num 4, 2005:387-400.
Ouziri M., Verdier C., Flory A., “Semantic indexing for intelligent browsing of distributed
data”, 13th Int. Conf. On Information and Intelligent System, IIS 2003, June 2-5, 2003,
Zakopane, Poland.
Ouziri M., Verdier C., Flory A., “Data integration and user modelling: An approach based on
Topic Maps and Description Logics”, ICEIS 2005, 6th Int. Conf. on Enterprise
Information Systems, supported by ACM, Miami, USA, May 24-28, 2005.
Quantin C., Binquet C., Bourquard K., Pattisina R., Gouyon-Cornet B., Ferdynus C.;
Gouyon J.B., Allaert F.A., “Which are the best identifiers for record linkage?”, pp. 221-
227, (7) Medical Informatics and the Internet in Medicine, ISSN 1463-9238, Publisher:
Taylor and Francis Ltd, Vol. 29, Issues 3-4, Numbers 3-4/September-December 2004.
Ross K.A., Janevski A., “Querying faceted databases”, Second International Workshop on
Semantic Web and Databases, 29-30 August 2004, Co-located with VLDB 2004, Toronto,
Canada, 30 August - 3 September 2004.
Scherrer J.R., “The hospital information system integrated patient records”, Comput Methods
Programs Biomed, 1995 Sep-Oct, 48, (1-2), 1-6.
Sigel A., “Towards knowledge organization with Topic Maps”, XML Europe 2000, Palais des
congrès Paris, 12-16 juin 2000.
Sol E.J., “Ambience Intelligence with the Ubiquitous Network, the Embedded Computer
Devices and the Hidden Databases”, Invited paper, proceedings of the 27th International
Conference on Very Large Data Bases, Sept. 11-14, 2001, Roma, Italy.
Soula G., Delquie P., Vitton P., Bartoli J.M., Fieschi M., “Designing medical hypermedia :
the Forum project”, Stud Health Technol Inform, 1997, 43 Pt B, 712-6.
Verdier C. Flory A., « An information system for epidemiology based on a computer-based
medical record », Journal Methods of Information in Medicine, Vol. 33, 5/94, december
1994, pp. 496-501.
Widhalm R., Mueck T.A. “Merging Topics in Well-Formed XML Topic Maps”, Lecture
Notes in Computer Science, Publisher: Springer-Verlag GmbH, Vol. 2870/2003, pp. 64-
79.
Zhu H., Madnick S.E., Siegel M.D., Representation and reasoning about changing semantics
in data sources”, Second International Workshop on Semantic Web and Databases, 29-30
August 2004, Co-located with VLDB 2004, Toronto, Canada, 30 August - 3 September
2004.