Elisabeth Métais

Reliability of diagnostic coding in intensive care patients

IntroductionAdministrative coding of medical diagnoses in intensive care unit (ICU) patients is mandatory in order to create databases for use in epidemiological and economic studies. We assessed the reliability of coding between different ICU physicians.MethodOne hundred medical records selected randomly from 29,393 cases collected between 1998 and 2004 in the French multicenter Outcomerea ICU database were studied. Each record was sent to two senior physicians from independent ICUs who recoded the diagnoses using the International Statistical Classification of Diseases and Related Health Problems: Tenth Revision (ICD-10) after being trained according to guidelines developed by two French national intensive care medicine societies: the French Society of Intensive Care Medicine (SRLF) and the French Society of Anesthesiology and Intensive Care Medicine (SFAR). These codes were then compared with the original codes, which had been selected by the physician treating the patient. A specific comparison was done for the diagnoses of septicemia and shock (codes derived from A41 and R57, respectively).ResultsThe ICU physicians coded an average of 4.6 ± 3.0 (range 1 to 32) diagnoses per patient, with little agreement between the three coders. The primary diagnosis was matched by both external coders in 34% (95% confidence interval (CI) 25% to 43%) of cases, by only one in 35% (95% CI 26% to 44%) of cases, and by neither in 31% (95% CI 22% to 40%) of cases. Only 18% (95% CI 16% to 20%) of all codes were selected by all three coders. Similar results were obtained for the diagnoses of septicemia and/or shock.ConclusionIn a multicenter database designed primarily for epidemiological and cohort studies in ICU patients, the coding of medical diagnoses varied between different observers. This could limit the interpretation and validity of research and epidemiological programs using diagnoses as inclusion criteria.

Enhancing information systems management with natural language processing techniques

Natural language and databases are core components of information systems. They are related to each other because they share the same purpose: the conceptualization aspects of the real world in order to deal with them in some way. Natural language processing (NLP) techniques may substantially enhance most phases of the information system lifecycle, starting with requirements analysis, specification and validation, and going up to conflict resolution, result processing and presentation. Furthermore, natural language based query languages and user interfaces facilitate the access to information for anyone and allow for new paradigms in the usage of computerized services. This paper investigates the use of NLP techniques in the design phase of information systems. Then, it reports on data base querying and information retrieval enhanced with NLP.

The linguistic level: contribution for conceptual design, view integration, reuse and documentation

Abstract This article presents the linguistic level of KHEOPS, describing its structure and the information it contains. Even though natural language specification interfaces are a main aspect of the KHEOPS environment we limit our discussion to the paraphrasing, view integration, and schema reuse aspects, presenting their main characteristics and how they make use of the said level. In fact, the major problem in these procedures, is the need to understand the schema semantics. To help in this sense, the linguistic level of KHEOPS offers semantic relations, Fillmores semantic cases and Sowas conceptual graphs stored in semantic electronic dictionaries.

Using linguistic knowledge in view integration: toward a third generation of tools

Abstract This paper addresses the problem of view integration in a CASE tool environment which is aiming at the elaboration of a conceptual schema of an application. The previous integration tools were mainly based on syntax and structure comparisons. A new generation of intelligent tools is now arising, assuming that view integration algorithms must also capture the deep semantics of the objects represented in the views. Dealing with the semantics of the objects is now a realistic objective, thanks to the research results obtained in the natural language area. This paper presents the definition of a view integration algorithm enhanced by the use of linguistic knowledge. This algorithm mainly consists of a semantic unification of views which are described using an extended entity-relationship model. It is combined with natural language techniques such as Fillmores semantic cases and Sowas conceptual graphs, supported by semantic dictionaries.

Ontology-Based Data Cleaning

Multi-source information systems, such as data warehouses, are composed of a set of heterogeneous and distributed data sources. The relevant information is extracted from these sources, cleaned, transformed and then integrated. The confrontation of two different data sources may reveal different kinds of heterogeneities: at the intensional level, the conflicts are related to the structure of the data. At the extensional level, the conflicts are related to the instances of the data. The process of detecting and solving the conflicts at the extensional level is known as data cleaning. In this paper, we will focus on the problem of differences in terminologies and we propose a solution based on linguistic knowledge provided by a domain ontology. This approach is well suited for application domains with intensive classification of data such as medicine or pharmacology. The main idea is to automatically generate some correspondence assertions between instances of objects. The user can parametrize this generation process by defining a level of accuracy expressed using the domain ontology.

A design tool for object oriented databases

This paper describes a design methodology for an object oriented database, based on a semantic network. This approach is based on the assumption that semantic data models are more powerful and more easy to use than current proposed object oriented data models. They are especially more powerful in representing integrity constraints and various relationships. Object oriented data models are generally based only on class hierarchies and inheritance, plus their ability to represent the behaviour of objects. But this latter capability is generally provided through an algorithmic language which cannot be considered as a conceptual language. In this paper, we combine the two categories of data models and give a procedure on how to translate the conceptual model to the logical model.

Text Categorization for Multi-label Documents and Many Categories

In this paper, we propose a new classification method that addresses classification in multiple categories of textual documents. We call it Matrix Regression (MR) due to its resemblance to regression in a high dimensional space. Experiences on a medical corpus of hospital records to be classified by ICD (International Classification of Diseases) code demonstrate the validity of the MR approach. We compared MR with three frequently used algorithms in text categorization that are k-Nearest Neighbors, Centroide and Support Vector Machine. The experimental results show that our method outperforms them in both precision and time of classification.

Towards an Automatic Detection of Sensitive Information in a Database

In order to validate user requirements, tests are often conducted on real data. However, developments and tests are more and more outsourced, leading companies to provide external staff with real confidential data. A solution to this problem is known as Data Scrambling. Many algorithms aim at smartly replacing true data by false but realistic ones. However, nothing has been developed to automate the crucial task of the detection of the data to be scrambled. In this paper we propose an innovative approach - and its implementation as an expert system - to achieve the automatic detection of the candidate attributes for scrambling. Our approach is mainly based on semantic rules that determine which concepts have to be scrambled, and on a linguistic component that retrieves the attributes that semantically correspond to these concepts. Since attributes can not be considered independently from each other we also address the challenging problem of the propagation of the scrambling among the whole database. An important contribution of our approach is to provide a semantic modelling of sensitive data. This knowledge is made available through production rules, operationalizing the sensitive data detection

CITOM: An incremental construction of multilingual topic maps

This paper proposes the CITOM approach for an incremental construction of multilingual Topic Maps. Our main goal is to facilitate users navigation across documents available in different languages. Our approach takes into account three types of information sources: (a) a set of multilingual documents, (b) a domain thesaurus and (c) all the possible questioning sources such as FAQ and users or experts requests about documents. In this paper we present the different steps of the proposed approach to construct the Topic Map and the pruning process of the generated Topic Map. We validate our approach with a real corpus from the sustainable construction domain.

PersonLink: An Ontology Representing Family Relationships for the CAPTAIN MEMO Memory Prosthesis

In the context of the CAPTAIN MEMO memory prosthesis for elderly, we propose the PersonLink ontology for modeling, storing and reasoning on “family relationships” links. Rules are provided to infer new links and/or check inconsistencies in the inputs. On the one hand PersonLink is as generic as possible and is integrated in the linked data formalisms; on the other hand a prosthesis has to be adaptable to users. Thus the PersonLink ontology defines rigorously and precisely family relationships, and takes into account the differences that may exist between cultures/languages, including new relationships emerging in our societies nowadays. The transition from one culture/language to another one cannot be solved with a simple translation of terms, but refers to a meta-ontology and associated mechanisms.