Semantic integration in healthcare networks

Abstract

A seamless support of information flow for increasingly distributed healthcare processes requires to integrate heterogeneous IT systems into a comprehensive distributed information system. Different standards contribute to ease this integration. In a research project focussing on the development of a reference architecture for inter-institutional health information systems, we identified concurring standards currently in use. We therefore categorized these integration standards by distinguishing between technical and semantic integration on the one hand, and data and functional integration on the other hand. In addition, standards for semantic integration are roughly categorized according to their scope. By placing standards into a corresponding matrix a “semantic gap” is revealed, which cannot be covered by standards as it contains volatile medical concepts. As a conclusion, it is recommended to conceptually consider the necessity of system evolution in system architectures and also in future integration standards.

Introduction

Healthcare increasingly changes from isolated treatment episodes towards a continuous treatment process involving multiple healthcare professionals and various institutions. This change motivates comprehensive, inter-institutional IT support in health information systems and imposes new demanding requirements for IT [1]. IT applications should guide data acquisition in a way that data are placed in a meaningful context from the beginning, so that they are ready for reuse in different contexts without the need to manually index or transform the data. To achieve such an IT support, heterogeneous IT systems have to be integrated into a comprehensive distributed information system. Integrating autonomous software components, however, is a difficult task, as individual applications usually are not designed to cooperate. Applications are often based on differing conceptualizations of the application domain. Today powerful integration tools (e.g. application servers, object brokers, different kinds of message-oriented middleware, and workflow management systems [2]) are available to overcome technical and syntactical heterogeneity of autonomous system components. Yet, semantic heterogeneity remains as a major barrier to seamless integration of autonomously developed software components (cf. [3]). Semantic heterogeneity occurs when there is disagreement about the meaning, interpretation or intended use of the same or related data [4]. It occurs in different contexts, like database schema integration, ontology mapping, or integration of different terminologies. The underlying problems are more or less the same, though they are often complex and still poorly understood. Stonebraker characterizes disparate systems as “islands of information” and points out two major factors which aggravate systems integration [5]:

• Each island (i.e. application) will have its own meaning of enterprise objects.

• Each island will have data that overlaps data in other islands. This partial redundancy generates a serious data integrity problem.

Based on this statement, data integration can be led back to a mapping problem (how to map different conceptualizations in a semantically correct way) and a synchronization problem (how to ensure mutual consistency of redundant data which are stored in different databases under the control of autonomous applications). The mapping problem is essentially related to the schema integration problem of database systems, which has been extensively discussed in the database literature in recent years (e.g. [6], [7], [8], [9]). A major perception in data integration research has been that schema integration cannot be automated in general. In [10] it is stated: “The general problem of schema integration is undecidable.” Heiler states that “understanding data and software can never be fully automated” [11]. As a consequence, the process of schema integration always needs a human integrator for certain semantic decisions. Colomb even goes a step further by stating that there are cases where no consistent interpretation of heterogeneous sources is possible (“fundamental semantic heterogeneity”) [12]. In such cases one either has to accept a low degree of data quality, or systems have to be modified to resolve fundamental semantic heterogeneity.

In order to reduce the integration efforts caused by semantic heterogeneity standards for systems integration are needed. Moreover, as medicine is a rapidly evolving domain, concepts for system evolution are needed. Fortunately, there are already far reaching standards that support information interchange in the medical domain. Yet, healthcare software is still far away from plug and play compatibility, and systems integration is typically a difficult process. In a research project in which we focus on the development of a reference architecture for comprehensive information systems in healthcare networks [1], [13], we have identified concurring and semantically overlapping standards. To get an overview of the standards’ characteristics and interrelations, we have arranged them to a system of standards which we find to be helpful for architecture development.