An object-oriented design for automated navigation of semantic networks inside a medical data dictionary

Abstract

In this paper we present a data dictionary server for the automated navigation of information sources. The underlying knowledge is represented within a medical data dictionary. The mapping between medical terms and information sources is based on a semantic network. The key aspect of implementing the dictionary server is how to represent the semantic network in a way that is easier to navigate and to operate, i.e. how to abstract the semantic network and to represent it in memory for various operations. This paper describes an object-oriented design based on Java that represents the semantic network in terms of a group of objects. A node and its relationships to its neighbors are encapsulated in one object. Based on such a representation model, several operations have been implemented. They comprise the extraction of parts of the semantic network which can be reached from a given node as well as finding all paths between a start node and a predefined destination node. This solution is independent of any given layout of the semantic structure. Therefore the module, called Gießen Data Dictionary Server can act independent of a specific clinical information system. The dictionary server will be used to present clinical information, e.g. treatment guidelines or drug information sources to the clinician in an appropriate working context. The server is invoked from clinical documentation applications which contain an infobutton. Automated navigation will guide the user to all the information relevant to her/his topic, which is currently available inside our closed clinical network.

Introduction

Modern hospital information systems (HIS) strive to comprise a complete electronic patient record. They can however no longer be considered as singular applications. Van de Velde [38] defines a modern hospital information system with the following words:

An HIS serves essentially as a medium for communication between the diverse collaborating functional subsystems or units in a hospital. It acts therefore as skeleton to facilitate the integration of various systems operating at subsystem level.

In addition to patient data such information systems will offer access to an increasing variety of medical information sources. Such sources can comprise medical textbooks and guidelines as well as literature search programs such as MEDLINE. They can provide full library access, drug data sources, teaching programs and research databases. In some institutions the integration of medical contents has been formalized, e.g. in the American IAIMS (integrated advanced information management systems) program [17]. However, with more and more medical information sources available, information searching becomes increasingly tedious and cannot be performed quickly during daily routine any more. Approaches have been proposed in the shape of so called infobuttons [15], [11]. The generic idea behind this approach is that data from an application the user is working with, are fed into a search mechanism which narrows the search results to the most appropriate information sources in the given circumstance.

Considering van de Velde’s definition of an HIS as an integration skeleton it becomes clear that such infobutton mechanisms must be independent from any specific HIS application. Only then they can be integrated in many different existing or future HIS applications dealing with patient data. Furthermore, infobutton mechanisms should also be independent of any specific information source or guideline. They should rather be able to deal with as many different kinds of information sources as possible. Within this paper, we present navigation mechanisms and an implemented solution based on a medical data dictionary as means to achieve those goals. We coined the expression ‘context sensitive information presentation’ for this solution which allows us to follow links to different appropriate information sources based on terms derived from the HIS application [5].

In our case the clinical application must be merely complemented with an infobutton. Clinical staff, which record e.g. clinical findings or intensive care plans on a computer, can press this infobutton to retrieve information for a clinical term they have marked in their application. Such clinical terms may comprise, e.g. medical findings, medical events or procedures. The infobutton activates an Internet browser and sends the term of interest to the Gießen data dictionary server (GDDS). Receiving the search term the dictionary server starts searching the semantic network of the data dictionary and returns the semantically related website information to the Internet browser.

Gießen University hospital is a 1300 bed facility caring for about 38 000 admissions and 300 000 outpatient treatments per year. The Gießen University Hospital Information System is a comprehensive and continuously evolving HIS with its roots dating back to 1989. A central clinical database, maintained on a Tandem mainframe, still forms the backbone and contains most of the clinical patient data [35]. The information system itself has been converted to a client server architecture integrating a variety of commercial and non-commercial departmental applications as described by van de Velde [27]. A total of 2000 clinical workstations and printers spread throughout the hospital in wards, doctors’ offices, outpatient clinics, theatres and partially at bedsides have access to the HIS. Starting in 1993, a variety of commercial information sources such as MEDLINE, clinical textbooks, etc. as well as in-house compiled information sources called the electronic books have been made available on clinical workstations as separate applications [1], [8], [32], [33]. The concept has always been that all required clinical information sources should be made available on each single workstation, rendering possible what we termed one stop information shopping [32]. Many information sources have been converted to HTML (hypertext markup language), the presentation format used also in the Internet and an intranet has been established with Web browsers available on most clinical workstations.

To provide an independent mechanism for the linking of on-line information sources to clinical applications a medical data dictionary has been used. Medical data dictionaries can be defined as

A central thesaurus for the controlled definition of the medical vocabulary to be applied in an HIS, which is also capable to represent the semantic relationships existing between all HIS objects and to link the local vocabulary to standardized international nomenclatures and knowledge sources [34].

There are two common structures to arrange terms in medical data dictionaries (MDDs) [3], [13]. The terms in early data dictionaries are grouped in strict hierarchies [22], [26], [39]. More modern data dictionary concepts allow semantic links besides hierarchy [16], [23], overcoming the limitations imposed by purely hierarchic relationships. Another shortcoming of some elder data dictionaries was a limitation in the depth of a hierarchy and the number of items per hierarchy level [29].

Semantic structure is flexible and has no structural limitation. Medical terms can be linked with semantic relationships as needed. It allows for a simple and straightforward representation of real world conditions. Therefore it is comparatively easy to link any given medical term or procedure derived from clinical applications to any information source. On the other hand, navigating semantic networks is somewhat complicated. The method to represent the semantic network for navigation purposes should have no limitation either.

Gaining its first experience with the US derived PTXT (pointer to text) data dictionary [22] during an experimental HELP HIS installation, the staff of Gießen University medical information processing department has developed several of its own data dictionaries [3], [28], [30], [31]. The data dictionary used in our project is based on the architecture of MDD-GIPHARM [31] which was originally designed to support drug charting and the implementation of knowledge based functions to monitor the prescription [7]. The MDD-GIPHARM architecture allows the definition of a vocabulary of medical terms, which can be linked by semantic relationships. In the following sections whenever we reference MDD this architecture is meant.

To allow flexible linking between on-line information sources and different clinical applications an independent web based dictionary server has been implemented on top of the MDD. When passive data dictionaries can be described as a thesaurus which can be used merely for lookup purposes, a dictionary server can perform active duties offering some kind of application programming interface (API) to different applications. Therefore it is independent of any given information system. The concept of the active dictionary service is described in Refs. [4], [5]. We will reference our Gießen data dictionary server as GDDS.