Saoussen Sakji

Multiple Terminologies in a Health Portal: Automatic Indexing and Information Retrieval

Background: In the specific context of developing quality-controlled health gateways, several standards must be respected (e.g. Dublin Core for metadata element set; thesaurus MeSH as the controlled vocabulary to index Internet resources; HON code to accredit quality of health Web sites). These standards were applied to create the CISMeF Web site (French acronym for Catalog & Index of Health Internet resources in French). Objective: In this work, the strategic shift of the CISMeF team is intended to index and retrieve French resources not anymore with a single terminology (MeSH thesaurus) but with the main health terminologies available in French (ICD 10, SNOMED International, CCAM, ATC). Methods & Results: Since 2005, we have developed the French Multi-Terminology Indexer (F-MTI), using a multi-terminology approach and mappings between health terminologies. This tool is used for automatic indexing and information retrieval. Conclusion: Since the last quarter of 2008, F-MTI is daily used in the CISMeF production environment and is connected to a French Health Multi-Terminology Server.

Improving information retrieval with multiple health terminologies in a quality-controlled gateway.

BackgroundThe Catalog and Index of French-language Health Internet resources (CISMeF) is a quality-controlled health gateway, primarily for Web resources in French (n=89,751). Recently, we achieved a major improvement in the structure of the catalogue by setting-up multiple terminologies, based on twelve health terminologies available in French, to overcome the potential weakness of the MeSH thesaurus, which is the main and pivotal terminology we use for indexing and retrieval since 1995. The main aim of this study was to estimate the added-value of exploiting several terminologies and their semantic relationships to improve Web resource indexing and retrieval in CISMeF, in order to provide additional health resources which meet the users’ expectations.MethodsTwelve terminologies were integrated into the CISMeF information system to set up multiple-terminologies indexing and retrieval. The same sets of thirty queries were run: (i) by exploiting the hierarchical structure of the MeSH, and (ii) by exploiting the additional twelve terminologies and their semantic links. The two search modes were evaluated and compared.ResultsThe overall coverage of the multiple-terminologies search mode was improved by comparison to the coverage of using the MeSH (16,283 vs. 14,159) (+15%). These additional findings were estimated at 56.6% relevant results, 24.7% intermediate results and 18.7% irrelevant.ConclusionThe multiple-terminologies approach improved information retrieval. These results suggest that integrating additional health terminologies was able to improve recall. Since performing the study, 21 other terminologies have been added which should enable us to make broader studies in multiple-terminologies information retrieval.

Information retrieval in context using various health terminologies

Information retrieval is a branch of computer science concerned with the acquisition, storage, search and selection of information. From the user point of view, the access to information can be carried out in a deliberate way through an information retrieval system, or in a passive way through an information filtration system. CISMeF (Catalogue and Index of the French-speaking Medical Sites) is a health portal aiming to catalogue and index the most important French-speaking institutional health information sources in order to make them available to health professionals, medical students and the general public. The Internet resources were manually indexed and have remained mono-terminological from 1995 to 2007 originally based exclusively on the MeSH thesaurus (Medical Subject Headings). Categorization allows a contextual information retrieval parameterized according to the users needs. In 2007, the CISMeF team directed its objectives towards a multi-terminological universe by the integration of the medical data heterogeneous sources into its back-office. To date, the practical application is the creation of a bilingual (French/English) drug information portal in order to facilitate the user information retrieval about drugs.

Recherche d’information multiterminologique en contexte: étude préliminaire

Background: Information retrieval becomes more relevant thanks to document (semi)automatic indexation with terminologies adapted to the domain. Being interested in the medical domain, we used the main health terminologies available in French in two different contexts: (a) during the health resource indexation process and (b) during the information retrieval. Methods: Within the CISMeF catalog (French acronym of Catalogue et index des Sites Medicaux Francophones), we have evolved from a monoterminological world (based on the MeSH thesaurus) to a multi-terminological universe (based on ICDIO, CCAM, SNOMED terminologies and ATC classification). This new structure allows us to make the information retrieval in two different contexts: to retrieve (a) reports from the electronic health record (EHR) of patients and (b) resources from a health portal (such as CISMeF). Results: Both contexts share the same representation meta-model, however, they differ during the information retrieval, according to the size and the nature of the corpus, the role of dates and the query, which can include both structured and unstructured elements in EHR. Conclusion: In 2008, the information retrieval has already become multi-terminological in the Drug Information Portal and will become available in the catalog CISMeF at the beginning of 2009.

Évaluation de l’indexation des comptes rendus médicaux à l’aide d’un outil états-unien adapté pour le français

Objective: To evaluate the ability of the bilingual (French-English) language vocabulary server to identify the same main concepts of ICD-10 (International Classification of Disease v.10) in the same set of French and English Electronic Health Records. Material and methods: The Intelligent Natural Language Processor (iNLP) was adapted to parse French as well as English clinical content. Six clinical notes were translated from French to English. The English notes were processed with the English version of ICD-10 and the French notes were processed with the French version. The resultant codes were compared. Results: The English Version contained 36 concepts of which 26 were in common with the French Version (72.2%). Failure analysis showed that errors were mostly due to missing synonymy in the French ICD-10 terminology server and to problems associated with translating the records from French to English. Discussion: Multi-lingual iNLP is feasible and has the potential to enable querying across national boundaries and multi-lingual communities.

Un modèle de données adapté à la recherche d’information dans le dossier patient informatisé: Étude, conception et évaluation

Background: Electronic Health Records (EHRs) have an important role in medical decision making and are considered as being a major part of medical support system. To ensure this role effectively, the information must be easily found, even in voluminous and large number of EHRs. This requires to develop search capabilities for EHR information retrieval (IR). Methods: To perform this, we propose a data model adapted to IR. The data analysis of EHRs of the Rouen University Hospital led us to consider them as being a set of informational elements linked by temporal and conceptual relationships. This model consists of two main entities to manage the information elements and their metadata. After implementation, we have evaluated the capacity of the proposed model to take into account all data from the EHRs and its adaptation to IR. Results: We have performed 25 queries on EHR (20 correspond to single patient and 5 to multi-patients). The results in 19 cases (including 3 partial results) were evaluated as successful. These results confirm the ability of the model to take into account most of relevant data of EHR in IR and clearly show the differences in complexity of queries between the initial and our model of IR. Conclusion: The preliminary evaluation of the data model we have developed has shown its adaptation to IR in EHR. Nevertheless, further work on larger sets is needed to confirm this.

Relevance of Google-customized search engine vs. CISMeF quality-controlled health gateway.

CISMeF (acronym for Catalog and Index of French Language Health Resources on the Internet) is a quality-controlled health gateway conceived to catalog and index the most important and quality-controlled sources of institutional health information in French. The goal of this study is to compare the relevance of results provided by this gateway from a small set of documents selected and described by human experts to those provided by a search engine from a large set of automatically indexed and ranked resources. The Google-Customized search engine (CSE) was used. The evaluation was made using the 10th first results of 15 queries and two blinded physician evaluators. There was no significant difference between the relevance of information retrieval in CISMeF and Google CSE. In conclusion, automatic indexing does not lead to lower relevance than a manual MeSH indexing and may help to cope with the increasing number of references to be indexed in a controlled health quality gateway.