VR-PROJECT 2003-2005

THE PROBLEM OF SEMANTIC INTEROPERABILITY -
APPLICATIONS WITHIN HEALTH CARE

Hans Åhlfeldt hans.ahlfeldt@imt.liu.se
Daniel Karlsson daniel.karlsson@imt.liu.se
Håkan Petersson hakan.petersson@imt.liu.se
Mikael Nyström mikael.nystrom@imt.liu.se

Department of Biomedical Engineering / Medical Informatics
Linköpings universitet
S-581 83 Linköping

In cooperation with the REFTERM-project, Karolinska institutet, Stockholm:
Gunnar Klein gunnar@klein.se
Gunnar Nilsson gunnar.nilsson@nlpo.sll.se

It is well known that the health care system is faced with a series of challenges concerning quality and cost-effectiveness. The distribution of cost-effective health care allowing the patient to take active part in the caring process, provision of evidence-based care on all levels in the system and effective use and reuse of information are key issues for the health care organization. The information and communication technology infrastructure should therefore reflect the view of the health care system as a seamless system where information can flow, although under strict regulation, across organizational and professional borders. The complex application area of distributed health care provide in this perspective a series of general research problems which we believe must be studied to allow successful systems to be implemented. A list of such related research areas is presented below. The areas share the basic problem of semantic interoperability, which simply means that semantics is preserved in communication between health care actors using information systems, a condition which should be natural but has proven to be very hard to achieve.

Terminological systems used in health care are systems supporting communication of information through standardized use of language. These systems include thesauri, nomenclatures, classifications, vocabularies and also formally and non-formally represented health-care models. Traditionally the systems have been disseminated through paper-based publications. Now however, the integration of terminological systems with health care information systems is becoming more viable. Thus, the focus of this research application is on terminological systems within health care and their use in health care information systems such as documenting and retrieval systems.

Consistent use of language, or more generally any piece of information used for communication purposes, is a prerequisite for high-quality classification and for semantic interoperability. An ideal terminological system should thus serve as an inter-lingua in communication in-between health care workers, and permit data comparison across organizational, professional, and even national borders. Concept representation systems based on description logic (such as OpenGALEN, FaCT or KL-ONE) is a key technology , allowing formal manipulation and interpretation. The health care domain is, as mentioned above, rich of terminological systems targeted towards different areas and specialties, and several models are proposed as basic reference terminology, possibly providing a common framework for further integration and mapping between specific terminological systems.

The documentation of clinically useful, patient-specific information is fundamental for management of health care related problems. Over the last years, the structure and content of a multi-professional patient record have gained much interest, one reason being the introduction of electronic patient records (EPRs), and another being organizational issues originating from the objective of using the patient record for communication over organizational and professional boarders and as an active instrument in planning and outcome analysis. The EPR has to face the requirements of quality assurance and evidence-based practice, but at the same time preserve the richness of the patient-centered story. In this perspective, it is well known that the traditional paper-based patient record suffers from a series of limitations.

Consequently, demands of information handling within the health care sector range from clinically useful, patient-specific information to a variety of aggregation levels for follow-up and statistical reporting. Several coding systems are for this purpose put into use in domains such as diagnoses, health problems, interventions, and procedures. The challenge is to assure high information quality on different levels of abstraction, and to allow aggregation according to different aspects or viewpoints.

Considerable progress has been made in the field of text processing and information retrieval during the last decenium but these achievements have not yet reached a level of maturity or reached the health care sector as ready-to-use products. Successful information retrieval rely on consistent use of an indexing language which, depending on the actual domain, consists of terms from any of the terminological systems.

Information retrieval techniques include mathematical modeling of the relation between document content, which in our context is the patient record, and the end-user query. Generally, the set of terms used for indexing and thereby also for retrieval is a flat list without internal structure apart from what is given lexically. An interesting research perspective is how the structure of conceptual models imposed on the indexing system could be utilized to improve the performance of search engines. Our objective is to develop methods for quantifying the semantic distance between index terms based on concept models including generic, part-whole and other associative relations.

The development of effective techniques for information retrieval will be necessary for realization of decision and information support, especially so in a distributed environment with a mix of information systems and different end-users.

The objective is, through analysis and experiments, to explore the above mentioned problem areas. The research program is divided in three tracks suitable for three PhD thesis work. The research challenges are as described above intertwined, allowing the PhD students to gain knowledge and skills upon a common ground, at the same time allowing them to specialize on their own sub-area. The result of the work will be synthesized into two major applications, both of high relevance for the health care organization (see appendix I with description of significance for research area and the health care sector for further details).

The objectives of the research program are the following:

• To transform traditionally organized coding systems into concept representation systems allowing formal concept classification including subsumption and equivalence testing. Further to analyze useful dimensions in the concept representation space and to explore and evaluate methods and tools for support of the coding process . The subject under study will be a series of terminological systems and a class of terminology browsers facilitating browsing, selection and cross-referencing. (PhD track 1: Terminology issues in health care)
• To analyze the relation between information quality and level of detail in the use of coding systems including statistical description of inter-rater variability in ordinal as well as nominal ratings. Further to analyze the content of (selected parts of) the multi-professional patient record in relation to the rubrics in relevant coding systems, thereby increasing the knowledge concerning the potential impact of the structure and content of coding systems on the documentation of health related problems in the patient record. (PhD track 2: Abstraction and information quality )
• To explore methods for indexing of patient records with categories from controlled terminologies including analysis of measures of semantic distance between entities in ordinal and nominal terminologies. Further to develop and test methods for information retrieval including integration of methods from mathematical modeling and text retrieval and resource description formats used in the www-framework. (PhD track 3: Automated indexing and information retrieval).
• To design and to do prototype implementation of terminology services in the so called ‘information platform’ developed jointly by two county councils in the south east region (Application 1: Terminology services)
• To design, implement and evaluate tools to generate user-specific views of information in the information platform – multi-views of patient information stored in a mix of health care information systems presented to different users (patients, next-of-kin, health care professionals etc.). (Application 2: View generator for patient information)