- Email: [email protected]
ARIANE: integration of information databases within a hospital intranet
International Journal of Medical Informatics 49 (1998) 297 – 309
ARIANE: integration of information databases within a hospital intranet Michel Joubert a,b,*, Sylvain Aymard a, Dominique Fieschi a, Franc¸oise Volot a,b, Pascal Staccini c, Jean-Jacques Robert d, Marius Fieschi a,b b
a LERTIM, Faculte´ de Me´decine, Uni6ersite´ de la Me´diterrane´e, Marseille, France Ser6ice de l’Information Me´dicale, Hoˆpital de la Timone — Adultes, 254, rue Saint Pierre, 13385 Marseille Cedex 5, France c De´partement d’Information Me´dicale, Hoˆpital Saint Roch, Nice, France d EURITIS, Technopoˆle de Chaˆteau-Gombert, Marseille, France
Received 25 February 1998; received in revised form 15 March 1998; accepted 28 March 1998
Abstract Large information systems handle massive volume of data stored in heterogeneous sources. Each server has its own model of representation of concepts with regard to its aims. One of the main problems end-users encounter when accessing different servers is to match their own viewpoint on biomedical concepts with the various representations that are made in the databases servers. The aim of the project ARIANE is to provide end-users with easy-to-use and natural means to access and query heterogeneous information databases. The objectives of this research work consist in building a conceptual interface by means of the Internet technology inside an enterprise Intranet and to propose a method to realize it. This method is based on the knowledge sources provided by the Unified Medical Language System (UMLS) project of the US National Library of Medicine. Experiments concern queries to three different information servers: PubMed, a Medline server of the NLM; The´riaque, a French database on drugs implemented in the Hospital Intranet; and a Web site dedicated to Internet resources in gastroenterology and nutrition, located at the Faculty of Medicine of Nice (France). Accessing to each of these servers is different according to the kind of information delivered and according to the technology used to query it. Dealing with health care professional workstation, the authors introduced in the ARIANE project quality criteria in order to attempt a homogeneous and efficient way to build a query system able to be integrated in existing information systems and to integrate existing and new information sources. © 1998 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Information systems; Information retrieval; Unified Medical Language System; Health care workstation; Databases; Internet
* Corresponding author. Tel.: +33 491387084; fax: +33 491385749; e-mail: [email protected] 1386-5056/98/$19.00 © 1998 Elsevier Science Ireland Ltd. All rights reserved. PII S1386-5056(98)00084-7
298
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
1. Introduction Large information systems, such as Hospital Information Systems, handle massive volume of data stored in heterogeneous sources. This has serious consequences, a lot of data is manually processed, some information is unprocessed and costs are high. A first solution for improving these systems is the re-engineering of their software components with recent computer technics and architectures, client/server, object-orientation, communication standards and so on. This approach satisfies the constraints of heterogeneity of hardware and software and guarantees the interoperability of the services inside a system. It can be qualified as a ‘syntactical’ solution. A second solution, which does not exclude modern computer technology, consists in the total revision of the systems components according to a knowledge engineering approach. It is an answer to the semantic heterogeneity of data processed by the software components of the information systems. It proposes a semantic information integration that is necessary for applications to operate in a corporate environment. This ‘semantic’ approach emphasizes the notion of conceptual analysis and design [1]. The various servers in a computerized information system implement data according to the way they represent medical concepts. Each server has its own model of concepts representation with regard to its aims. One of the main problems encountered by end-users when accessing different servers inside a unique environment is to match their own viewpoint on concepts with the various representations that are made in the databases servers. The aim of the project ARIANE is to provide end-users with easy-to-use and natural means to access and query heterogeneous information databases [2,3]. A few years ago the International Medical Informatics Associ-
ation (IMIA) organized a workshop on ‘the health care professional workstation’. Among the recommendations that emerge from this conference some are of special interest today [4]: Utilize component-based architectures to define, design and develop the health professional workstation. Advocate a distributed, client/server model using open systems and international, national, or industry standards. Extend the component-based architecture of health care information systems to regional, national and international dimensions using emerging communication technologies. Develop and promote the acceptance of common vocabularies for health care professionals to facilitate the sharing of health care information and knowledge. Promote evaluation studies of health care workstation environments, including criteria for quality, access, cost and satisfaction. These conference results set forth the vision of a health care infrastructure in which the workstation acts as an enabler, giving professionals access to information when, where and how it is needed. All the above mentioned issues are an integral part of the project ARIANE objectives. Semantic approaches are based on ontologies which provide a representational vocabulary for a given domain with a set of rules that constrain the meaning of the terms sufficiently to enable consistent interpretation of data framed in that vocabulary [5]. The Unified Medical Language System (UMLS) project of the US National Library of Medicine [6] includes in its knowledge sources (Semantic Network (SN) [7], Metathesaurus [8], Specialist Lexicon [9]) a part of the medical knowledge and, above all, almost of the biomedical vocabularies. The
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
UMLS can be considered today as an operational and suitable ontology for the biomedical domain. In previous research works, the authors have shown how to use the UMLS knowledge sources with the intent to query information databases. The model the authors developed [10,11] and implemented [12] to express and process queries to information databases [13], is based on the conceptual graphs formalism [14]. To achieve the goal, the authors investigated another component of the UMLS, the Information Sources Map (ISM) intended to route users to various identified information servers [15]. Even if the authors did not access servers identified by the NLM, the present study used the data structure and the principle of the ISM to declare the servers they accessed by way of experiment. They are: PubMed [16], a currently free of charge server of Medline at the NLM site; The´riaque, a database on all the drugs available in French hospitals the authors implemented into the University Hospital Intranet [17]; and a Web site located in the Faculty of Medicine of Nice (France) which registers information and delivers access to other sites in the field of gastroenterology and nutrition [18]. Access to each of these servers is different with regard to the kind of information it delivers and to the technology used to query it.
299
bases. Thus, an Intranet can be designed that respects these constraints. Moreover, it can be implemented following a method in such a way that end-users cannot be disorientated by unorganized data and information structures [20,21]. One of the means to achieve this goal is to index all the pages according to a unique vocabulary, such as MeSH, for instance. Nowadays, this is possible since MeSH is a world-wide recognized thesaurus, available for bibliographical servers and relational databases servers as well and since the Internet technology standards recommend to use keywords in META tags in the heading of HTML pages [22]. The drawing of Fig. 1 illustrates the objectives of the project ARIANE. This diagram is widely inspired by two ambitious projects intended to help end-users to retrieve information in large, heterogeneous and dynamically modified environments, Intelligent Integration of Information [23] and InfoSleuth [24]. With regard to the studys concern,
2. Methods Authors have shown the inherent difficulties they may have to organize clinical information in an Internet environment [19]. Thus, the Internet does not provide today the means necessary to integrate heterogeneous information databases in a semantic way. At the opposite, inside a given enterprise it is desirable to have rules and constraints for building, maintaining and accessing data-
Fig. 1. General architecture of the ARIANE prototype.
300
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
it appears that health care professionals often need access information in databases. Typically, they are today in relation with a computer system composed of their own workstation connected to the enterprise network and by the way of a larger network, such as the Internet, to other distant sites. Either they know the information server, its location and the way to query the related database, or they need a ‘mediation layer’ able to give them access to information in a possibly seamless way. In the first case they click on a button or icon displayed in a toolbar on the screen. Such a case is typically a client application installed on their workstation that runs a session with another application running on a server. In large enterprises that cover thousands of professional workstations, it does not seem really practical today to think to install and to maintain various client applications connected to different information sources for information retrieval process only. In the second case, end-users query a service delivered by a server inside the enterprise network. This latter service can be viewed as a ‘broker’ between a client, an end-user and providers, the information sources. The broker ‘knows’ that information sources exist, how to access them, but does not know generally how to query them. Thus it needs access ‘mediators’ which are capable to initiate a dialog between end-users and information servers. In the example of Fig. 1, mediators are able to communicate with servers such as a documentary server (e.g. Medline), a relational database management system (e.g. The´riaque, a French database on drugs) and a hypertext site (e.g. HepGen a French Web site dedicated to Internet resources in gastroenterology and nutrition). The major part of the studys objectives consists today in building a conceptual interface by means of the Internet technology
inside a hospital Intranet and to propose a method to carry it out. This means that the physical interface between end-users and databases is organized according to the semantics provided by an ontology. Since the UMLS SN offers a sufficiently explicit set of semantic relationships between types of concepts, the method the authors propose is to organize the access according to this knowledge representation. Moreover, the Metathesaurus provides with the current largest biomedical vocabulary. Thus, both the SN and the Metathesaurus can be considered as the foundations to implement an effective conceptual interface between end-users and the above broker. As demonstrated in above mentioned research works, an end-user’s query is formalized by a conceptual graph that represents the deep structure of a natural language sentence. For instance, if the current interest of an end-user is the treatment of gastric ulcers by ranitidine, the built graph is: [Pharmacological substance: ranitidine] (treats) [Disease or syndrome: gastric ulcer] where ‘Pharmacological substance’ and ‘Disease or syndrome’ are semantic types of the SN to which ranitidine and gastric ulcer are connected respectively in the UMLS knowledge sources. For a broker to work it must have two inputs: a precise qualification of the current query; and a knowledge of the available information sources. The Metathesaurus is able to qualify precisely a query by the mean of subheadings when using MeSH keywords. Pursuing the above example, a piece of declarative knowledge issued from the MeSH syntax and semantics can serve to translate the semantic relation ‘treats’ instantiated between ranitidine and gastric ulcer to qualify both as follows:
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
[ranitidine]/therapeutic use, pharmacokinetics, administration and dosage, pharmacology [gastric ulcer]/drug therapy This knowledge is not present in the UMLS knowledge sources and must be built with this purpose. It establishes the correspondence for each couple of semantic types and for each relevant semantic relationship possibly instantiated (this latter information is delivered by the SN), between the subheadings that can be assigned and the two concepts connected to their respective types and involved in a given relation. At this step of the process, the broker has analyzed a query. Then it is in a position to scan a catalog of its information providers. The ISM allows the registration of all the information servers declared in the Intranet. Moreover, the ISM describes by the means of MeSH keywords, sometimes qualified by subheadings, the expected content of information sources. The result of a matching process between the query and these definitions operates the selection of the target servers. Now, the broker is able to propose to the end-user a list of possible and coherent information servers. When an end-user selects one of these target information servers, the broker activates the right mediator in order to dialog effectively with the related provider of information. This mediator is activated by a message that contains the data necessary to initiate the communication with the related server. The role of a mediator is more syntactical than the semantic work of the broker. A mediator translates the transmitted query into the syntax of the server query language. In order to achieve its task it may need to find information in two different sources: the Metathesaurus for the needs of synonymy or multilingual applications; and the ISM to
301
obtain complementary information related to the target server. It may contain also the knowledge about information servers, for instance, the correct syntax for writing a query (e.g. how to query PubMed), or the correct way to query the index engine of a Web site (e.g. Excite on the HepGen Web site). At this step, two cases may arise according to the architecture depicted by the diagram of Fig. 1, either a communication is initiated between an end-user and a provider of information, or a dialog is established between an information server and its corresponding mediator. Since the authors think that the first case is the most general and the easiest to implement today with regard to the objectives, even if it is the most difficult to model, the authors will present in the next section experiments done in this frame. But, the authors know there are also information servers that end-users can not refer interactively at any moment, such as archiving systems, PACS for instance. In this latter case a mediator must be able to complete its task by initiating a communication protocol between the information server and the end-user, or itself as an intermediary. The authors do not consider this eventuality in the present framework.
3. Results In order to illustrate the above principles, the authors developed a stand-alone application that implements access to the three mentioned types of information servers. Let’s remember that a particular conceptual interface has been implemented for the sake of this prototype. A more sophisticated and integrated interface should be implemented later. Fig. 2 illustrates how this conceptual interface works.
302
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Fig. 2. A scenario for the conceptual interface.
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Step (1): a user types a set of words intended to search a concept in the UMLS vocabulary. Step (2): the conceptual interface displays a list of concepts whose names contain all these words. With a mouse click, the user selects on the screen the concept he/she wishes. Step (3): the conceptual interface displays on the screen all the contexts of use in which the selected concept occurs. In case of a concept issued from MeSH, for instance, this is materialized by the display of the hierarchies in which it occurs. Each hierarchy focuses on the context and shows its ancestors, its children when they are and its siblings. At this time the user selects an occurrence of the concept. These steps use the Metathesaurus knowledge only. Steps (4), (5) and (6) allow the user to operate similar selections for another concept. The conceptual interface ‘stored’ two concepts qualified by their respective contexts of use. It can now connect each of them to their respective types in the SN and then select by itself the semantic relationships that can apply between them. Step (7): the conceptual interface proposes to the user a list of possible semantic relationships. When he/she selects one of these relationships, the user builds an elementary query such as presented in the previous section: ‘gastric ulcer treated by ranitidine’. Control is now given to the broker. It exploits the query to search relevant available resources to route the user’s request towards appropriate information servers. In order to achieve its aim it needs information contained in two knowledge sources: A declarative knowledge used to translate a semantic relationship between concepts into MeSH subheadings associated to these concepts. Translation depends on the
303
respective types of the associated concepts. For instance, the relation ‘treats’ between a disease and a drug will yield to association of the subheading ‘drug treatment’ to the disease. In some cases, the translation is more precise. For instance, the relation ‘treats’ introduces four subheadings when it is linked to a pharmacologic substance, ‘therapeutic use’ and ‘administration & dosage’ represent practical notions, when ‘pharmacokinetics’ and ‘pharmacology’ represent more theoretical notions in therapeutics. After processing, the initial query is translated into a list of keywords qualified by MeSH subheadings. The second input the broker exploits is the ISM. In the ISM, each server is described by different fields such as: its name; its language; a textual description; a protocol of communication; and so on. In many cases, the description of a server is completed with a list of MeSH keywords and subheadings that indicate the kind of information it delivers. A matching process between the representation of the query and the description of the servers allows the broker to select in the ISM a list of possible relevant sources and to propose it to the user. Matching is operated according to the MeSH structure, so that information in relation with ‘gastric ulcer’ may be found in an information source dedicated to ‘gastroenterology diseases’. The broker does not eliminate from the set of candidates the servers whose descriptions match the query partially only. For instance, a large database on drugs is indexed in the ISM by keywords in relation with pharmacological substances and not in relation to pathologies necessarily. Nevertheless it may be considered as a candidate since a pharmacological substance occurs in the query. Remember that if the broker exploits the Metathesaurus, it does not use the SN.
304
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
The broker is now able to propose to the user a list of information servers that can possibly deliver the information he/she wishes. Fig. 3 summarizes the following access to information: Screen dump (1): the broker displays on the screen a list of servers that are able to deliver information in relation with the user’s query. Each server is pointed by a hypertext link and shortly presented by a textual description issued from the ISM. A mouse click on any of these links activates a mediator which formulates an appropriate query to the target server. Three of them are detailed below. Screen dump (2): a SQL mediator has been experimented with The´riaque, a database on drugs implemented in the Hospital Intranet. The mediator exploits scripts that have been implemented with the database whose intent are to give access to information according to various access points. In the current case, the mediator activates a script for searching a pharmacological substance and gives ‘ranitidine’ as a parameter. The server The´riaque then displays a page dynamically fulfilled with the information contained in its database. Following a hypertext way, links give then access to the commercial products that contain this substance and so on. Screen dump (3): a hypertext mediator has been experimented with HepGen, a Web site dedicated to gastroenterology and nutrition. The Web pages are static pages indexed according the standard recommendations. Keywords have been chosen in MeSH. The mediator activates a searcher implemented on the related Web site. In this case ‘gastric ulcer’ only has been retained to query the server. It returns a page that proposes a list of links to pages indexed according to the query. A mouse click on a link allows the user to visit a page
and then to navigate inside the site and to visit other Web sites following hypertext links. Screen dump (4): a bibliographical mediator has been experimented with PubMed, a NLM search service that gives access to databases such as Medline. The internal PubMed query language is close to the HTML syntax. The mediator translates the initial query into this language and activates at the appropriate URL the needed CGI with the related string of characters. The return is the page that PubMed would display in response to a user querying directly the service. Here, the major benefit for end-users is certainly to add automatically MeSH subheadings to qualify the concepts involved in their queries. Remember that each mediator uses shortcuts that allows users to access information directly without being obliged to: connect the server; visit a home page; fulfill fields in a search page; and so on. When no result is returned by an information server, a mediator then activates the home page of the related server. Then the user knows that wished information resides possibly here, but he/she has to find it by him/herself. This application has been developed following the current Microsoft Internet/Intranet technology. End-users access the conceptual interface through a standard Internet browser. The conceptual interface and the broker are made of a set of Active Server Pages which contain Visual Basic scripts. The mediators are Visual Basic scripts. The whole is running under Windows NT and MS-Internet Server.
4. Discussion A few years ago, the Internet and World Wide Web technology emerged and has been exploited successfully by medical informatics researchers. Some efforts are intended to
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Fig. 3. A scenario for the broker and the mediators.
305
306
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
provide end-users with easy-to-use and universally available access to information databases [25], when other works allow the integration of various kinds of biomedical information [26]. Nowadays, it seems not realistic to conceive a unique method to implement web sites on the Internet. Moreover, it would be unnatural presently. But, in the case of an internal Web, an Intranet, which is developed with the intent to increase the facilities of documentation and the capabilities of communication and to implement a collaborative work, rules of design and rules of implementation are necessary [27]. As felt by its designers themselves, the UMLS may be an effective base for building ‘better’ interfaces [28]. The method the authors propose uses the UMLS knowledge sources as an ontology that provides with a framework for building a homogeneous way to query heterogeneous databases in an Intranet. In the studys architecture a broker exploits such a query to identify the possible servers to address. When selected, the destination to the server is processed by an appropriate mediator that knows how to translate part of the request into the query language of the related server. The authors experimented this method with proprietary query languages, standard query languages such as SQL and hypertextbased searchers on Web sites. Translation of queries expressed as conceptual graphs into the query languages of information databases seems easy, even if the conceptual structure of the query is not kept in most cases. In case of a documentary server, such as Medline, the conceptual structure serves to qualify semantically a query to the server. In case of relational databases, only pieces of the conceptual queries are retained. A Web site is a special case of queried information database. If it is designed following the same principles that allowed to implement the conceptual interface, this means, if its pages are indexed
with keywords picked out from a widely recognized thesaurus, then it could be homogeneously integrated within a coherent Intranet, even if it is distant. Nevertheless, the mechanism of surfing over the Web does not guarantee homogeneity since following a hypertext link from a page may direct an end-user towards pages in other sites that are built according to another method. The authors think that the method they illustrated respects mostly the recommendations issued from the IMIA conference above mentioned. Nevertheless, the authors also attempted to manage the developments with the integration of criteria of quality for health care workstation environments, related to information access and users’ satisfaction. Because the authors want to integrate efficient information retrieval services within a professional health care workstation, the authors have to anticipate the benefits for endusers and to evaluate the impact that such a system may have on them. This question includes different aspects which have been previously described [29]: Homogeneity in and between searches and results. Homogeneity of results is the ability of a system to provide a user with results having the same presentation independently of their origin. Homogeneity in searches comes from the principle of having a same way to request whatever the databases concerned are. Reuse and predictability are linked in the sense that a same set of functions gives a consistent result whatever the different environments are. And it is the way this system is. Therefore the user is able to predict the type of results he is going to get. Consistency means that any action provides a user with coherent results. Consistency is linked with reuse and homogeneity in a sense that the more a system requires
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
standardized ways of querying and presents standardized results, the easier the user will learn how to use and reuse it in order to obtain coherent answers. The following two points to take also into account are more sociological than the former ones. They concern the interactivity endusers may have with a system [30]: The number of actions to be done before accessing to the information requested. It is linked to the user interface because the faster a user masters the system the best and the quickest he/she will get the results. As users well know how to use the system, they are then able to reduce the number of actions to carry on. Adaptation which takes into account the amount of effort a user has to do to be able to initially understand and consequently use a system in an efficient way. In ARIANE project, because the principle is to facilitate the way a user can defines his/her request and because a same standard browser is used to display and query, the effort to be able to fully use the system is well reduced. These points are tightly connected one to each other. Heterogeneity of sources may entail inconsistency. Consistency and coherence entail reusability and predictability. Finally attempting efficient integration of new sources of information leads to manage an ‘impact cycle’ starting from homogeneity of search and adaptation and going back to homogeneity of results. Authors who experimented integration of services such as decision support, literature searching and Web exploration using the Metathesaurus [31], conclude to the necessity to develop an sharable representation of information that could be used to determine what type of information an end-user might need during interaction. With the intent to integrate heterogeneous information sources
307
the authors are faced to the following challenge: either build a dedicated mediator able to communicate with each new source of information; or either try to design classes of mediators according to the kind of information sources they are able to address. The authors have experimented the first solution, but it is not realistic to think they will follow this way when the number of sources will increase. The second solution entails that the authors would be able to define what an information source is in the frame of the information retrieval objectives: what it is; what it contains; how to access it; and so on. This means that the authors would be able to design a mediator dedicated to information retrieval in any bibliographical system (a librarian?), a mediator able to formulate a SQL query to any relational database whatever it contains (a computer engineer?), a mediator able to guide an end-user in the pages of a Web site (a webmaster, perhaps!). This means the authors would be able to model what an information source is. Only at this time they would be able to design and implement enhanced mediators able to dialog effectively with existing information sources and capable to recognize easily new plugged sources. Where resides the benefit of having efficient mediators if the broker is not? How could we model information sources without modeling information? According to the above considerations again, the authors need to model the knowledge exploited by the broker to enlarge its capabilities. So, the authors come back to one of the primitive objectives of the project ARIANE that is to model medical concepts, in the framework of an information retrieval process. Thus, the authors close the above impact cycle, if a broker is able to treat queries well, if it addresses efficient mediators that know information servers well, then the authors can expect to decrease the heterogeneity of the
308
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
returns due to the heterogeneity of the servers, to increase the consistency and the coherence of the results and finally to improve the end-users adaptation to such a system.
[8]
Acknowledgements
[9]
This work has been partially founded by the French Ministry of Education and Research. The authors thank the US National Library of Medicine which gracefully provided them with the results of the UMLS project. This work has been achieved with the 1997 release. The authors thank the ‘Assistance Publique—Hoˆpitaux de Marseille’ and the ‘Faculte´ de Me´decine de Nice’ which created favorable conditions for this work. The database on drugs The´riaque is developed and maintained by the ‘Centre National d’Information sur le Me´dicament Hospitalier’ (CNIMH).
References [1] J.F. Sowa, Conceptual analysis for knowledgebase design, Methods Inform. Med. 34 (1995) 165–171. [2] M. Joubert, J.J. Robert, F. Miton, M. Fieschi, The Project ARIANE: Conceptual Queries to Information Databases. In: J.J. Cimino (Ed.), JAMIA Symposium Suppl. 1996, pp. 378–382. [3] M. Joubert, M. Fieschi, J.J. Robert, et al., UMLSbased conceptual queries to biomedical information databases—an overview of the project ARIANE. To appear in JAMIA. [4] M. Ball, J. Silva, J. Douglas, et al., Special issue: the health care professional workstation, Int. J. Biomed. Comput. 34 (1994). [5] T.R. Gruber, A translation approach to portable ontologies, Knowl. Acquis. 5 (1993) 199– 220. [6] D.A.B. Lindberg, B.L. Humphreys, A.T. McCray, The unified medical language system, Methods Inform. Med. 32 (1993) 281–291. [7] A.T. McCray, The UMLS Semantic Network. In: L.C. Kingsland (Ed.), Proceedings of the Thir-
[10]
[11]
[12]
[13]
[14]
[15]
[16] [17]
[18] [19]
[20]
[21]
teenth SCAMC. IEEE Computer Society Press 1989, pp. 503 – 507. M.S. Tuttle, D.D. Sheretz, M.S. Erlbaum, et al., Implemeting Meta-1: the First Version of the UMLS Metathesaurus. In: L.C. Kingsland (Ed.), Proceedings of the Thirteenth SCAMC. IEEE Computer Society Press 1989, pp. 483 – 487. A.T. McCray, A.R. Aronson, A.C. Browne, et al., UMLS knowledge for biomedical language processing, Bull. Med. Libr. Assoc. 81 (1993) 184– 194. M. Joubert, M. Fieschi, J.J. Robert, A Conceptual Model for Information Retrieval with UMLS. In: C. Safran (Ed.), Proceedings of the Seventeenth SCAMC. McGraw-Hill, New York, 1993, pp. 715 – 719. M. Joubert, F. Miton, M. Fieschi, J.J. Robert, A Conceptual Graphs Modelling of UMLS Components. In: R. Greenes, H. Peterson, D. Protti (Eds.), Proceedings of MEDINFO 95, North-Holland, Amsterdam, 1995, pp. 90 – 94. J.J. Robert, M. Joubert, L. Nal, M. Fieschi, A Computational Model of Information Retrieval with UMLS. In: J. Ozbolt (Ed.), JAMIA Symposium Suppl. 1994, pp. 167 – 71. M. Joubert, M. Fieschi, J.J. Robert, A. Tafazzoli, Users conceptual views on medical information databases, Int. J. Biomed. Comput. 37 (1994) 93– 104. J.F. Sowa, Conceptual Structures: Information Processing in Mind and Machine, Addison Wesley, Reading, MA, 1984. D.R. Masys, B.L. Humphreys, Structure and Function of the UMLS Information Sources Map. In: K.C. Lun, P. Degoulet, T.E. Piemme, et al., (Eds.), Proceedings of MEDINFO ‘92. North-Holland, Amsterdam, 1992, pp. 1518 – 1521. PubMed. http://www.ncbi.nlm.nih.gov/PubMed. M. Franc¸ois, M. Joubert, D. Fieschi, M. Fieschi, Implementation of a Database on Drugs into a University Hospital Intranet. Proceedings of MEDINFO ‘98. To appear. HepGen. http://noe.unice.fr/hepgen.nice. W.R. Hersh, K.E. Brown, L.C. Donohoe, et al., Cliniweb: managing clinical information on the world wide web, JAMIA 96 (3) 273 – 280. F. Volot, M. Joubert, D. Fieschi, M. Fieschi, A UMLS-based Method for Integrating Information Databases into an Intranet. In: D. Masys (Ed.), JAMIA Symposium Suppl. 1997, pp. 495 – 499. M. Joubert, F. Volot, D. Fieschi, M. Fieschi, Conceptual Integration of Information Databases
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
[22] [23] [24]
[25]
[26]
into an Intranet. Proceedings of MEDINFO ‘98. To appear. World Wide Web Consortium. http://www.w3.org/ Intelligent Integration of Information. http://dc.isx.com/I3. R. Bayardo, W. Bohrer, R. Brice, et al., Semantic Integration of Information in Open and Dynamic Environments. In: Proceedings of ACM SIGMOD ‘97. 1997, pp. 192–206. C.G. Chute, D.L. Crowson, J.D. Buntrock, Medical Information Retrieval and WWW Browsers at Mayo. In: R.D. Gardner (Ed.), JAMIA Symposium Suppl. 1995, pp. 903–907. W.M. Detmer, E.H. Shortliffe, A Model of Clinical Query Management that Supports Integration of Biomedical Information over the World Wide Web. In: R.D. Gardner (Ed.), JAMIA Symposium Suppl. 1995, pp. 898–902.
.
309
[27] R. Bernard, The Corporate Intranet: create and manage an internal web for your organization. Wiley Computer, 1996. [28] D.A.B. Lindberg, B.L. Humphreys, The UMLS Knowledge Sources: tools for building better user interfaces. In: R.A. Miller, (Ed.), Proceedings of the Fourteenth SCAMC. IEEE Computer Society Press 1990, pp. 121 – 125. [29] F. Garzotto, L. Malnetti, P. Paolini, Hypermedia design, analysis and evaluation issues, Comm. ACM 38 (8) (1995) 74 – 86. [30] P. Bernoux, La Sociologie des Organisations-initiation the´orique suivie de 12 cas pratiques. Editions du Seuil 1985. [31] W.M. Detmer, G.O. Barnett, W.R. Hersh, MedWeaver: Integrating Decision Support, Literature Sarching and Web Exploration using the UMLS Metathesaurus. In: D. Masys (Ed.), JAMIA Symposium Suppl. 1997, pp. 490 – 494.
ARIANE: integration of information databases within a hospital intranet Michel Joubert a,b,*, Sylvain Aymard a, Dominique Fieschi a, Franc¸oise Volot a,b, Pascal Staccini c, Jean-Jacques Robert d, Marius Fieschi a,b b
a LERTIM, Faculte´ de Me´decine, Uni6ersite´ de la Me´diterrane´e, Marseille, France Ser6ice de l’Information Me´dicale, Hoˆpital de la Timone — Adultes, 254, rue Saint Pierre, 13385 Marseille Cedex 5, France c De´partement d’Information Me´dicale, Hoˆpital Saint Roch, Nice, France d EURITIS, Technopoˆle de Chaˆteau-Gombert, Marseille, France
Received 25 February 1998; received in revised form 15 March 1998; accepted 28 March 1998
Abstract Large information systems handle massive volume of data stored in heterogeneous sources. Each server has its own model of representation of concepts with regard to its aims. One of the main problems end-users encounter when accessing different servers is to match their own viewpoint on biomedical concepts with the various representations that are made in the databases servers. The aim of the project ARIANE is to provide end-users with easy-to-use and natural means to access and query heterogeneous information databases. The objectives of this research work consist in building a conceptual interface by means of the Internet technology inside an enterprise Intranet and to propose a method to realize it. This method is based on the knowledge sources provided by the Unified Medical Language System (UMLS) project of the US National Library of Medicine. Experiments concern queries to three different information servers: PubMed, a Medline server of the NLM; The´riaque, a French database on drugs implemented in the Hospital Intranet; and a Web site dedicated to Internet resources in gastroenterology and nutrition, located at the Faculty of Medicine of Nice (France). Accessing to each of these servers is different according to the kind of information delivered and according to the technology used to query it. Dealing with health care professional workstation, the authors introduced in the ARIANE project quality criteria in order to attempt a homogeneous and efficient way to build a query system able to be integrated in existing information systems and to integrate existing and new information sources. © 1998 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Information systems; Information retrieval; Unified Medical Language System; Health care workstation; Databases; Internet
* Corresponding author. Tel.: +33 491387084; fax: +33 491385749; e-mail: [email protected] 1386-5056/98/$19.00 © 1998 Elsevier Science Ireland Ltd. All rights reserved. PII S1386-5056(98)00084-7
298
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
1. Introduction Large information systems, such as Hospital Information Systems, handle massive volume of data stored in heterogeneous sources. This has serious consequences, a lot of data is manually processed, some information is unprocessed and costs are high. A first solution for improving these systems is the re-engineering of their software components with recent computer technics and architectures, client/server, object-orientation, communication standards and so on. This approach satisfies the constraints of heterogeneity of hardware and software and guarantees the interoperability of the services inside a system. It can be qualified as a ‘syntactical’ solution. A second solution, which does not exclude modern computer technology, consists in the total revision of the systems components according to a knowledge engineering approach. It is an answer to the semantic heterogeneity of data processed by the software components of the information systems. It proposes a semantic information integration that is necessary for applications to operate in a corporate environment. This ‘semantic’ approach emphasizes the notion of conceptual analysis and design [1]. The various servers in a computerized information system implement data according to the way they represent medical concepts. Each server has its own model of concepts representation with regard to its aims. One of the main problems encountered by end-users when accessing different servers inside a unique environment is to match their own viewpoint on concepts with the various representations that are made in the databases servers. The aim of the project ARIANE is to provide end-users with easy-to-use and natural means to access and query heterogeneous information databases [2,3]. A few years ago the International Medical Informatics Associ-
ation (IMIA) organized a workshop on ‘the health care professional workstation’. Among the recommendations that emerge from this conference some are of special interest today [4]: Utilize component-based architectures to define, design and develop the health professional workstation. Advocate a distributed, client/server model using open systems and international, national, or industry standards. Extend the component-based architecture of health care information systems to regional, national and international dimensions using emerging communication technologies. Develop and promote the acceptance of common vocabularies for health care professionals to facilitate the sharing of health care information and knowledge. Promote evaluation studies of health care workstation environments, including criteria for quality, access, cost and satisfaction. These conference results set forth the vision of a health care infrastructure in which the workstation acts as an enabler, giving professionals access to information when, where and how it is needed. All the above mentioned issues are an integral part of the project ARIANE objectives. Semantic approaches are based on ontologies which provide a representational vocabulary for a given domain with a set of rules that constrain the meaning of the terms sufficiently to enable consistent interpretation of data framed in that vocabulary [5]. The Unified Medical Language System (UMLS) project of the US National Library of Medicine [6] includes in its knowledge sources (Semantic Network (SN) [7], Metathesaurus [8], Specialist Lexicon [9]) a part of the medical knowledge and, above all, almost of the biomedical vocabularies. The
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
UMLS can be considered today as an operational and suitable ontology for the biomedical domain. In previous research works, the authors have shown how to use the UMLS knowledge sources with the intent to query information databases. The model the authors developed [10,11] and implemented [12] to express and process queries to information databases [13], is based on the conceptual graphs formalism [14]. To achieve the goal, the authors investigated another component of the UMLS, the Information Sources Map (ISM) intended to route users to various identified information servers [15]. Even if the authors did not access servers identified by the NLM, the present study used the data structure and the principle of the ISM to declare the servers they accessed by way of experiment. They are: PubMed [16], a currently free of charge server of Medline at the NLM site; The´riaque, a database on all the drugs available in French hospitals the authors implemented into the University Hospital Intranet [17]; and a Web site located in the Faculty of Medicine of Nice (France) which registers information and delivers access to other sites in the field of gastroenterology and nutrition [18]. Access to each of these servers is different with regard to the kind of information it delivers and to the technology used to query it.
299
bases. Thus, an Intranet can be designed that respects these constraints. Moreover, it can be implemented following a method in such a way that end-users cannot be disorientated by unorganized data and information structures [20,21]. One of the means to achieve this goal is to index all the pages according to a unique vocabulary, such as MeSH, for instance. Nowadays, this is possible since MeSH is a world-wide recognized thesaurus, available for bibliographical servers and relational databases servers as well and since the Internet technology standards recommend to use keywords in META tags in the heading of HTML pages [22]. The drawing of Fig. 1 illustrates the objectives of the project ARIANE. This diagram is widely inspired by two ambitious projects intended to help end-users to retrieve information in large, heterogeneous and dynamically modified environments, Intelligent Integration of Information [23] and InfoSleuth [24]. With regard to the studys concern,
2. Methods Authors have shown the inherent difficulties they may have to organize clinical information in an Internet environment [19]. Thus, the Internet does not provide today the means necessary to integrate heterogeneous information databases in a semantic way. At the opposite, inside a given enterprise it is desirable to have rules and constraints for building, maintaining and accessing data-
Fig. 1. General architecture of the ARIANE prototype.
300
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
it appears that health care professionals often need access information in databases. Typically, they are today in relation with a computer system composed of their own workstation connected to the enterprise network and by the way of a larger network, such as the Internet, to other distant sites. Either they know the information server, its location and the way to query the related database, or they need a ‘mediation layer’ able to give them access to information in a possibly seamless way. In the first case they click on a button or icon displayed in a toolbar on the screen. Such a case is typically a client application installed on their workstation that runs a session with another application running on a server. In large enterprises that cover thousands of professional workstations, it does not seem really practical today to think to install and to maintain various client applications connected to different information sources for information retrieval process only. In the second case, end-users query a service delivered by a server inside the enterprise network. This latter service can be viewed as a ‘broker’ between a client, an end-user and providers, the information sources. The broker ‘knows’ that information sources exist, how to access them, but does not know generally how to query them. Thus it needs access ‘mediators’ which are capable to initiate a dialog between end-users and information servers. In the example of Fig. 1, mediators are able to communicate with servers such as a documentary server (e.g. Medline), a relational database management system (e.g. The´riaque, a French database on drugs) and a hypertext site (e.g. HepGen a French Web site dedicated to Internet resources in gastroenterology and nutrition). The major part of the studys objectives consists today in building a conceptual interface by means of the Internet technology
inside a hospital Intranet and to propose a method to carry it out. This means that the physical interface between end-users and databases is organized according to the semantics provided by an ontology. Since the UMLS SN offers a sufficiently explicit set of semantic relationships between types of concepts, the method the authors propose is to organize the access according to this knowledge representation. Moreover, the Metathesaurus provides with the current largest biomedical vocabulary. Thus, both the SN and the Metathesaurus can be considered as the foundations to implement an effective conceptual interface between end-users and the above broker. As demonstrated in above mentioned research works, an end-user’s query is formalized by a conceptual graph that represents the deep structure of a natural language sentence. For instance, if the current interest of an end-user is the treatment of gastric ulcers by ranitidine, the built graph is: [Pharmacological substance: ranitidine] (treats) [Disease or syndrome: gastric ulcer] where ‘Pharmacological substance’ and ‘Disease or syndrome’ are semantic types of the SN to which ranitidine and gastric ulcer are connected respectively in the UMLS knowledge sources. For a broker to work it must have two inputs: a precise qualification of the current query; and a knowledge of the available information sources. The Metathesaurus is able to qualify precisely a query by the mean of subheadings when using MeSH keywords. Pursuing the above example, a piece of declarative knowledge issued from the MeSH syntax and semantics can serve to translate the semantic relation ‘treats’ instantiated between ranitidine and gastric ulcer to qualify both as follows:
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
[ranitidine]/therapeutic use, pharmacokinetics, administration and dosage, pharmacology [gastric ulcer]/drug therapy This knowledge is not present in the UMLS knowledge sources and must be built with this purpose. It establishes the correspondence for each couple of semantic types and for each relevant semantic relationship possibly instantiated (this latter information is delivered by the SN), between the subheadings that can be assigned and the two concepts connected to their respective types and involved in a given relation. At this step of the process, the broker has analyzed a query. Then it is in a position to scan a catalog of its information providers. The ISM allows the registration of all the information servers declared in the Intranet. Moreover, the ISM describes by the means of MeSH keywords, sometimes qualified by subheadings, the expected content of information sources. The result of a matching process between the query and these definitions operates the selection of the target servers. Now, the broker is able to propose to the end-user a list of possible and coherent information servers. When an end-user selects one of these target information servers, the broker activates the right mediator in order to dialog effectively with the related provider of information. This mediator is activated by a message that contains the data necessary to initiate the communication with the related server. The role of a mediator is more syntactical than the semantic work of the broker. A mediator translates the transmitted query into the syntax of the server query language. In order to achieve its task it may need to find information in two different sources: the Metathesaurus for the needs of synonymy or multilingual applications; and the ISM to
301
obtain complementary information related to the target server. It may contain also the knowledge about information servers, for instance, the correct syntax for writing a query (e.g. how to query PubMed), or the correct way to query the index engine of a Web site (e.g. Excite on the HepGen Web site). At this step, two cases may arise according to the architecture depicted by the diagram of Fig. 1, either a communication is initiated between an end-user and a provider of information, or a dialog is established between an information server and its corresponding mediator. Since the authors think that the first case is the most general and the easiest to implement today with regard to the objectives, even if it is the most difficult to model, the authors will present in the next section experiments done in this frame. But, the authors know there are also information servers that end-users can not refer interactively at any moment, such as archiving systems, PACS for instance. In this latter case a mediator must be able to complete its task by initiating a communication protocol between the information server and the end-user, or itself as an intermediary. The authors do not consider this eventuality in the present framework.
3. Results In order to illustrate the above principles, the authors developed a stand-alone application that implements access to the three mentioned types of information servers. Let’s remember that a particular conceptual interface has been implemented for the sake of this prototype. A more sophisticated and integrated interface should be implemented later. Fig. 2 illustrates how this conceptual interface works.
302
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Fig. 2. A scenario for the conceptual interface.
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Step (1): a user types a set of words intended to search a concept in the UMLS vocabulary. Step (2): the conceptual interface displays a list of concepts whose names contain all these words. With a mouse click, the user selects on the screen the concept he/she wishes. Step (3): the conceptual interface displays on the screen all the contexts of use in which the selected concept occurs. In case of a concept issued from MeSH, for instance, this is materialized by the display of the hierarchies in which it occurs. Each hierarchy focuses on the context and shows its ancestors, its children when they are and its siblings. At this time the user selects an occurrence of the concept. These steps use the Metathesaurus knowledge only. Steps (4), (5) and (6) allow the user to operate similar selections for another concept. The conceptual interface ‘stored’ two concepts qualified by their respective contexts of use. It can now connect each of them to their respective types in the SN and then select by itself the semantic relationships that can apply between them. Step (7): the conceptual interface proposes to the user a list of possible semantic relationships. When he/she selects one of these relationships, the user builds an elementary query such as presented in the previous section: ‘gastric ulcer treated by ranitidine’. Control is now given to the broker. It exploits the query to search relevant available resources to route the user’s request towards appropriate information servers. In order to achieve its aim it needs information contained in two knowledge sources: A declarative knowledge used to translate a semantic relationship between concepts into MeSH subheadings associated to these concepts. Translation depends on the
303
respective types of the associated concepts. For instance, the relation ‘treats’ between a disease and a drug will yield to association of the subheading ‘drug treatment’ to the disease. In some cases, the translation is more precise. For instance, the relation ‘treats’ introduces four subheadings when it is linked to a pharmacologic substance, ‘therapeutic use’ and ‘administration & dosage’ represent practical notions, when ‘pharmacokinetics’ and ‘pharmacology’ represent more theoretical notions in therapeutics. After processing, the initial query is translated into a list of keywords qualified by MeSH subheadings. The second input the broker exploits is the ISM. In the ISM, each server is described by different fields such as: its name; its language; a textual description; a protocol of communication; and so on. In many cases, the description of a server is completed with a list of MeSH keywords and subheadings that indicate the kind of information it delivers. A matching process between the representation of the query and the description of the servers allows the broker to select in the ISM a list of possible relevant sources and to propose it to the user. Matching is operated according to the MeSH structure, so that information in relation with ‘gastric ulcer’ may be found in an information source dedicated to ‘gastroenterology diseases’. The broker does not eliminate from the set of candidates the servers whose descriptions match the query partially only. For instance, a large database on drugs is indexed in the ISM by keywords in relation with pharmacological substances and not in relation to pathologies necessarily. Nevertheless it may be considered as a candidate since a pharmacological substance occurs in the query. Remember that if the broker exploits the Metathesaurus, it does not use the SN.
304
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
The broker is now able to propose to the user a list of information servers that can possibly deliver the information he/she wishes. Fig. 3 summarizes the following access to information: Screen dump (1): the broker displays on the screen a list of servers that are able to deliver information in relation with the user’s query. Each server is pointed by a hypertext link and shortly presented by a textual description issued from the ISM. A mouse click on any of these links activates a mediator which formulates an appropriate query to the target server. Three of them are detailed below. Screen dump (2): a SQL mediator has been experimented with The´riaque, a database on drugs implemented in the Hospital Intranet. The mediator exploits scripts that have been implemented with the database whose intent are to give access to information according to various access points. In the current case, the mediator activates a script for searching a pharmacological substance and gives ‘ranitidine’ as a parameter. The server The´riaque then displays a page dynamically fulfilled with the information contained in its database. Following a hypertext way, links give then access to the commercial products that contain this substance and so on. Screen dump (3): a hypertext mediator has been experimented with HepGen, a Web site dedicated to gastroenterology and nutrition. The Web pages are static pages indexed according the standard recommendations. Keywords have been chosen in MeSH. The mediator activates a searcher implemented on the related Web site. In this case ‘gastric ulcer’ only has been retained to query the server. It returns a page that proposes a list of links to pages indexed according to the query. A mouse click on a link allows the user to visit a page
and then to navigate inside the site and to visit other Web sites following hypertext links. Screen dump (4): a bibliographical mediator has been experimented with PubMed, a NLM search service that gives access to databases such as Medline. The internal PubMed query language is close to the HTML syntax. The mediator translates the initial query into this language and activates at the appropriate URL the needed CGI with the related string of characters. The return is the page that PubMed would display in response to a user querying directly the service. Here, the major benefit for end-users is certainly to add automatically MeSH subheadings to qualify the concepts involved in their queries. Remember that each mediator uses shortcuts that allows users to access information directly without being obliged to: connect the server; visit a home page; fulfill fields in a search page; and so on. When no result is returned by an information server, a mediator then activates the home page of the related server. Then the user knows that wished information resides possibly here, but he/she has to find it by him/herself. This application has been developed following the current Microsoft Internet/Intranet technology. End-users access the conceptual interface through a standard Internet browser. The conceptual interface and the broker are made of a set of Active Server Pages which contain Visual Basic scripts. The mediators are Visual Basic scripts. The whole is running under Windows NT and MS-Internet Server.
4. Discussion A few years ago, the Internet and World Wide Web technology emerged and has been exploited successfully by medical informatics researchers. Some efforts are intended to
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
Fig. 3. A scenario for the broker and the mediators.
305
306
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
provide end-users with easy-to-use and universally available access to information databases [25], when other works allow the integration of various kinds of biomedical information [26]. Nowadays, it seems not realistic to conceive a unique method to implement web sites on the Internet. Moreover, it would be unnatural presently. But, in the case of an internal Web, an Intranet, which is developed with the intent to increase the facilities of documentation and the capabilities of communication and to implement a collaborative work, rules of design and rules of implementation are necessary [27]. As felt by its designers themselves, the UMLS may be an effective base for building ‘better’ interfaces [28]. The method the authors propose uses the UMLS knowledge sources as an ontology that provides with a framework for building a homogeneous way to query heterogeneous databases in an Intranet. In the studys architecture a broker exploits such a query to identify the possible servers to address. When selected, the destination to the server is processed by an appropriate mediator that knows how to translate part of the request into the query language of the related server. The authors experimented this method with proprietary query languages, standard query languages such as SQL and hypertextbased searchers on Web sites. Translation of queries expressed as conceptual graphs into the query languages of information databases seems easy, even if the conceptual structure of the query is not kept in most cases. In case of a documentary server, such as Medline, the conceptual structure serves to qualify semantically a query to the server. In case of relational databases, only pieces of the conceptual queries are retained. A Web site is a special case of queried information database. If it is designed following the same principles that allowed to implement the conceptual interface, this means, if its pages are indexed
with keywords picked out from a widely recognized thesaurus, then it could be homogeneously integrated within a coherent Intranet, even if it is distant. Nevertheless, the mechanism of surfing over the Web does not guarantee homogeneity since following a hypertext link from a page may direct an end-user towards pages in other sites that are built according to another method. The authors think that the method they illustrated respects mostly the recommendations issued from the IMIA conference above mentioned. Nevertheless, the authors also attempted to manage the developments with the integration of criteria of quality for health care workstation environments, related to information access and users’ satisfaction. Because the authors want to integrate efficient information retrieval services within a professional health care workstation, the authors have to anticipate the benefits for endusers and to evaluate the impact that such a system may have on them. This question includes different aspects which have been previously described [29]: Homogeneity in and between searches and results. Homogeneity of results is the ability of a system to provide a user with results having the same presentation independently of their origin. Homogeneity in searches comes from the principle of having a same way to request whatever the databases concerned are. Reuse and predictability are linked in the sense that a same set of functions gives a consistent result whatever the different environments are. And it is the way this system is. Therefore the user is able to predict the type of results he is going to get. Consistency means that any action provides a user with coherent results. Consistency is linked with reuse and homogeneity in a sense that the more a system requires
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
standardized ways of querying and presents standardized results, the easier the user will learn how to use and reuse it in order to obtain coherent answers. The following two points to take also into account are more sociological than the former ones. They concern the interactivity endusers may have with a system [30]: The number of actions to be done before accessing to the information requested. It is linked to the user interface because the faster a user masters the system the best and the quickest he/she will get the results. As users well know how to use the system, they are then able to reduce the number of actions to carry on. Adaptation which takes into account the amount of effort a user has to do to be able to initially understand and consequently use a system in an efficient way. In ARIANE project, because the principle is to facilitate the way a user can defines his/her request and because a same standard browser is used to display and query, the effort to be able to fully use the system is well reduced. These points are tightly connected one to each other. Heterogeneity of sources may entail inconsistency. Consistency and coherence entail reusability and predictability. Finally attempting efficient integration of new sources of information leads to manage an ‘impact cycle’ starting from homogeneity of search and adaptation and going back to homogeneity of results. Authors who experimented integration of services such as decision support, literature searching and Web exploration using the Metathesaurus [31], conclude to the necessity to develop an sharable representation of information that could be used to determine what type of information an end-user might need during interaction. With the intent to integrate heterogeneous information sources
307
the authors are faced to the following challenge: either build a dedicated mediator able to communicate with each new source of information; or either try to design classes of mediators according to the kind of information sources they are able to address. The authors have experimented the first solution, but it is not realistic to think they will follow this way when the number of sources will increase. The second solution entails that the authors would be able to define what an information source is in the frame of the information retrieval objectives: what it is; what it contains; how to access it; and so on. This means that the authors would be able to design a mediator dedicated to information retrieval in any bibliographical system (a librarian?), a mediator able to formulate a SQL query to any relational database whatever it contains (a computer engineer?), a mediator able to guide an end-user in the pages of a Web site (a webmaster, perhaps!). This means the authors would be able to model what an information source is. Only at this time they would be able to design and implement enhanced mediators able to dialog effectively with existing information sources and capable to recognize easily new plugged sources. Where resides the benefit of having efficient mediators if the broker is not? How could we model information sources without modeling information? According to the above considerations again, the authors need to model the knowledge exploited by the broker to enlarge its capabilities. So, the authors come back to one of the primitive objectives of the project ARIANE that is to model medical concepts, in the framework of an information retrieval process. Thus, the authors close the above impact cycle, if a broker is able to treat queries well, if it addresses efficient mediators that know information servers well, then the authors can expect to decrease the heterogeneity of the
308
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
returns due to the heterogeneity of the servers, to increase the consistency and the coherence of the results and finally to improve the end-users adaptation to such a system.
[8]
Acknowledgements
[9]
This work has been partially founded by the French Ministry of Education and Research. The authors thank the US National Library of Medicine which gracefully provided them with the results of the UMLS project. This work has been achieved with the 1997 release. The authors thank the ‘Assistance Publique—Hoˆpitaux de Marseille’ and the ‘Faculte´ de Me´decine de Nice’ which created favorable conditions for this work. The database on drugs The´riaque is developed and maintained by the ‘Centre National d’Information sur le Me´dicament Hospitalier’ (CNIMH).
References [1] J.F. Sowa, Conceptual analysis for knowledgebase design, Methods Inform. Med. 34 (1995) 165–171. [2] M. Joubert, J.J. Robert, F. Miton, M. Fieschi, The Project ARIANE: Conceptual Queries to Information Databases. In: J.J. Cimino (Ed.), JAMIA Symposium Suppl. 1996, pp. 378–382. [3] M. Joubert, M. Fieschi, J.J. Robert, et al., UMLSbased conceptual queries to biomedical information databases—an overview of the project ARIANE. To appear in JAMIA. [4] M. Ball, J. Silva, J. Douglas, et al., Special issue: the health care professional workstation, Int. J. Biomed. Comput. 34 (1994). [5] T.R. Gruber, A translation approach to portable ontologies, Knowl. Acquis. 5 (1993) 199– 220. [6] D.A.B. Lindberg, B.L. Humphreys, A.T. McCray, The unified medical language system, Methods Inform. Med. 32 (1993) 281–291. [7] A.T. McCray, The UMLS Semantic Network. In: L.C. Kingsland (Ed.), Proceedings of the Thir-
[10]
[11]
[12]
[13]
[14]
[15]
[16] [17]
[18] [19]
[20]
[21]
teenth SCAMC. IEEE Computer Society Press 1989, pp. 503 – 507. M.S. Tuttle, D.D. Sheretz, M.S. Erlbaum, et al., Implemeting Meta-1: the First Version of the UMLS Metathesaurus. In: L.C. Kingsland (Ed.), Proceedings of the Thirteenth SCAMC. IEEE Computer Society Press 1989, pp. 483 – 487. A.T. McCray, A.R. Aronson, A.C. Browne, et al., UMLS knowledge for biomedical language processing, Bull. Med. Libr. Assoc. 81 (1993) 184– 194. M. Joubert, M. Fieschi, J.J. Robert, A Conceptual Model for Information Retrieval with UMLS. In: C. Safran (Ed.), Proceedings of the Seventeenth SCAMC. McGraw-Hill, New York, 1993, pp. 715 – 719. M. Joubert, F. Miton, M. Fieschi, J.J. Robert, A Conceptual Graphs Modelling of UMLS Components. In: R. Greenes, H. Peterson, D. Protti (Eds.), Proceedings of MEDINFO 95, North-Holland, Amsterdam, 1995, pp. 90 – 94. J.J. Robert, M. Joubert, L. Nal, M. Fieschi, A Computational Model of Information Retrieval with UMLS. In: J. Ozbolt (Ed.), JAMIA Symposium Suppl. 1994, pp. 167 – 71. M. Joubert, M. Fieschi, J.J. Robert, A. Tafazzoli, Users conceptual views on medical information databases, Int. J. Biomed. Comput. 37 (1994) 93– 104. J.F. Sowa, Conceptual Structures: Information Processing in Mind and Machine, Addison Wesley, Reading, MA, 1984. D.R. Masys, B.L. Humphreys, Structure and Function of the UMLS Information Sources Map. In: K.C. Lun, P. Degoulet, T.E. Piemme, et al., (Eds.), Proceedings of MEDINFO ‘92. North-Holland, Amsterdam, 1992, pp. 1518 – 1521. PubMed. http://www.ncbi.nlm.nih.gov/PubMed. M. Franc¸ois, M. Joubert, D. Fieschi, M. Fieschi, Implementation of a Database on Drugs into a University Hospital Intranet. Proceedings of MEDINFO ‘98. To appear. HepGen. http://noe.unice.fr/hepgen.nice. W.R. Hersh, K.E. Brown, L.C. Donohoe, et al., Cliniweb: managing clinical information on the world wide web, JAMIA 96 (3) 273 – 280. F. Volot, M. Joubert, D. Fieschi, M. Fieschi, A UMLS-based Method for Integrating Information Databases into an Intranet. In: D. Masys (Ed.), JAMIA Symposium Suppl. 1997, pp. 495 – 499. M. Joubert, F. Volot, D. Fieschi, M. Fieschi, Conceptual Integration of Information Databases
M. Joubert et al. / International Journal of Medical Informatics 49 (1998) 297–309
[22] [23] [24]
[25]
[26]
into an Intranet. Proceedings of MEDINFO ‘98. To appear. World Wide Web Consortium. http://www.w3.org/ Intelligent Integration of Information. http://dc.isx.com/I3. R. Bayardo, W. Bohrer, R. Brice, et al., Semantic Integration of Information in Open and Dynamic Environments. In: Proceedings of ACM SIGMOD ‘97. 1997, pp. 192–206. C.G. Chute, D.L. Crowson, J.D. Buntrock, Medical Information Retrieval and WWW Browsers at Mayo. In: R.D. Gardner (Ed.), JAMIA Symposium Suppl. 1995, pp. 903–907. W.M. Detmer, E.H. Shortliffe, A Model of Clinical Query Management that Supports Integration of Biomedical Information over the World Wide Web. In: R.D. Gardner (Ed.), JAMIA Symposium Suppl. 1995, pp. 898–902.
.
309
[27] R. Bernard, The Corporate Intranet: create and manage an internal web for your organization. Wiley Computer, 1996. [28] D.A.B. Lindberg, B.L. Humphreys, The UMLS Knowledge Sources: tools for building better user interfaces. In: R.A. Miller, (Ed.), Proceedings of the Fourteenth SCAMC. IEEE Computer Society Press 1990, pp. 121 – 125. [29] F. Garzotto, L. Malnetti, P. Paolini, Hypermedia design, analysis and evaluation issues, Comm. ACM 38 (8) (1995) 74 – 86. [30] P. Bernoux, La Sociologie des Organisations-initiation the´orique suivie de 12 cas pratiques. Editions du Seuil 1985. [31] W.M. Detmer, G.O. Barnett, W.R. Hersh, MedWeaver: Integrating Decision Support, Literature Sarching and Web Exploration using the UMLS Metathesaurus. In: D. Masys (Ed.), JAMIA Symposium Suppl. 1997, pp. 490 – 494.