Fei Song, & William Soukoreff (1995). A Graphical Interface to a Semantic Medical Information System. Karp-95 Proceedings of the Second International Symposium on Knowledge Acquisition, Representation and Processing, pp. 107-109.

Graphical Interface to a Semantic Medical Information System

Fei Song and William Soukoreff

Dept. of Computing and Information Science
University of Guelph
Guelph, Ontario, Canada, N1G 2W1


Information technologies are having a profound impact on clinical practice and medical research. In particular, electronic medical records are changing the way institutions are administered as well as the way researchers work. With patient information stored on-line, doctors can readily access it over computer networks and researchers can retrieve relevant information at a speed much faster than searching through paper records. The focus of this paper is an electronic implementation of medical problem lists for the Ontario Veterinary College (OVC) at the University of Guelph. Problem lists can be seen as the core components of medical records, since they help organize other information such as examination results and treatment plans.

Problems and Problem Lists

As defined in Walker (1973), a problem can be anything that calls the attention of a doctor such as an illness, a medical condition, or any other condition that requires diagnosis or medical management or that interferes with the quality of a patient's life. In current practice at the OVC, problem descriptions are spread over several places in medical records: initial problems are stated in history and/or examination sections, current problems are stated in progress notes, and some of the problems are re-stated in the master problem list. The master problem list is supposed to record all the problems, but after going through all examinations and the initial writing, doctors often fail to copy them over to the master list. As a result, when a patient arrives for a re-check, doctors may have to look through many pages of the paper file in order to gather all the previously-stated problems. By storing problems on-line, each problem needs to be entered only once, since we can easily pull them together electronically into the master list. However, simply automating the paper records may not lend itself to an acceptable system, since the underlying linear organization of problems presents a number of serious drawbacks.

In a typical linear organization, each problem is identified with a unique number, followed by the fields of date, time, doctor, and a short description. The description is a free-format field, which allows a doctor to specify the name of the problem and its related properties, such as whether the problem is on-going or resolved, and the degree of uncertainty for an observation. One drawback of the linear organization is that deletions are not allowed; updates have to be handled by creating new entries and referencing the numbers of the previously mentioned descriptions. Another drawback is that it is difficult to see the logical relationships between problems, which are typically hierarchical and directly useful to doctors. Finally, with much old information cluttered and hierarchical relationships buried in a linear list, it can be time-consuming to construct a sub-list that shows the set of current problems and their relationships, since doctors may have to scan through all the pages of a patient's record.

The Semantic Model for Problem Lists

In reality, we found that doctors do not view each problem in isolation; they often intentionally relate initial problems to general problems, to even more general problems, and eventually, to diagnoses. This motived us to propose a semantic model that supports the following three useful features. The first feature is a taxonomy of problem classification that supports efficient retrieval of patient information. Each problem is labeled in terms of four axes: status (RESOLVED, ACTIVE, or INACTIVE), severity (MAJOR, MINOR, or INCONSEQUENTIAL), actions (BEING-TREATED or NOT-BEING-TREATED), and sureness (VERIFIED-CORRECT, LIKELY, UNLIKELY, or VERIFIED-INCORRECT). Using this classification, we can easily perform such search tasks as finding all the MAJOR and ACTIVE problems, finding all the BEING-TREATED problems, and the like. In addition, it should be possible to build a list of current problems that a patient is suffering from for a particular time. One possible crierion for such a list is to include all the problems that are UNRESOLVED (or ON-GOING) but are not VERIFIED-INCORRECT.

The second feature is a hierarchical organization of problems which indicates explicitly the relationships between problems. In a typical hierarchical structure, each node corresponds to a medical problem, and each link from left to right represents an abstraction relationship. At the lowest level on the left are the initial problems such as signs and symptoms. Higher levels of problems are added as doctors continue their investigation into the patient. Ideally, the rightmost-level nodes should be some general problems such as syndromes or diagnoses. Our investigation into the OVC domain reveals that the relationships between problems basically correspond to the various abstractions in a typical rule-based medical expert system. We can take an example from the well-known MYCIN domain to illustrate these abstractions: White-Blood-Cell < 2.5 indicates a low count (qualitative abstraction), which is a condition for Leukopenia (definitional abstraction). Leukopenia is a kind of immunosuppression (generalization), suggesting a Gram-negative infection (heuristic-classification), which may further suggest an infection of E. coli (refinement) after other details are also considered. A hierarchical organization is directly useful to doctors in that by making explicit the relationships between problems, we are able to capture the way a problem list is actually used in practice. A hierarchical organization is also flexible in that it reflects multiple levels of resolutions about the understanding of a patient. As a result, doctors will be able to describe their opinions at a natural level of abstraction, to an arbitrary level of detail, or even with the context of their observation.

The last feature of our model is a set of time labels for distinguishing different versions of a problem. Since a patient's conditions and a doctor's understanding often change over time, updates to problem descriptions are almost inevitable. However, for legal reasons, old problem descriptions cannot simply be deleted. Although a hierarchical organization helps to show the relationships between problems, the intuitive appeal will be reduced or even lost if it is cluttered with too much outdated information. Version graphs provide an economic way for representing time-related information. In a typical version graph, each problem is associated with an abstract version, which contains those time-independent properties. Other versions are time-dependent and are labeled by the dates and/or times they are created. All the changes are stored at individual problems and only changes are recorded in the new version of a problem. Based on the version graphs, we can easily construct the current problem hierarchy for a particular time (a dynamic entity called the snapshot). As a result, when displaying a snapshot, we can remove all the irrelevant information (either outdated or too new) to avoid any distraction to a doctor.

The Graphical Interface

An implementation of our semantic model needs to be built on an object-oriented representation, since the popular relational model does not allow explicit relationships between problem descriptions; nor does it support free-format text for the extended description of a problem and time labels for the different versions of a problem. Also, the implementation should provide a user-friendly interface for easy input and access to patient information, in order to compete with paper records. Current research has shown that computer record keeping is often not much quicker than that of paper records. As described in Nygren and Henriksson (1992), a paper record consists of a set of different documents in a specific order. These documents may have different shapes and can be marked with different signs and colors in the margins. As a result, it is possible to overview several pages at the same time, and to rapidly browse through a large number of pages. The speed an experienced user can achieve in "zooming-in" to the relevant parts is remarkable, and the amount of information covered by a glance is enormous.

Thus, a successful implementation of medical records should make the user interaction with the system as effortless as possible. We found that a graphical interface with supports for direct manipulation and full-view display is especially suitable for our application (Shneiderman, 1992). Such an interface allows a user to change search parameters by simply clicking the corresponding icons and see all the effects right away in a full-view display. It also allows the user to browse through the display by clicking each item to view its detailed explanation. One possible interface design may look like the following. On one panel, we could provide a set of icons for the major search parameters. They can be arranged into five lines. The first four lines correspond to the four axes of our taxonomy: status, severity, actions, and sureness. The last line is a double slide-bar, allowing us to set a time label for a particular year/month (finer distinctions can be supported by expanding such a mechanism). On another panel, we could provide the full-view display for the searched results, typically in the form of a hierarchical structure. The user can click each item in the display to see its detailed explanation or if the display is not over-complex, a brief description of all the items can be shown in another small panel. Different colors may also be used to distinguish certain major properties. Such an interface would allow a user to easily formulate a search request by directly selecting the corresponding icons and see the full-view display of the retrieved results immediately on the screen.

Current Status and Future Work

We are currently working on the design and implementation of a prototype to demonstrate the feasibility and acceptability of our semantic model. This involves collecting a reasonable set of cases, conducting user and task analysis, finishing an initial paper design, and having it implemented in a programming language. We have already collected over a dozen of cases in the domain of small animal cardiology, a representative subset in the OVC hospital. Detailed analysis of these cases shows that the representation of our semantic model is basically adequate. Considerable efforts are being made for the user and task analysis so that an effective graphical interface can be developed. The interface would allow the direct manipulation of a user and display all the relevant information in a full view. The prototype will be implemented on a personal computer (PC) in the Visual C++ language. The choice of PC is to increase the accessibility of the system so that it can be made widely available in the OVC hospital. The Visual C++ is an ideal tool for our implementation since it provides supports for both the object-oriented representation and the graphical interface. Our top-level goal is to develop an acceptable system to doctors, and the initial feedback from our domain experts and a small number of potential users has been positive and encouraging.

Our work may be expanded in several directions. We could still use the current model but extend the prototype to cover the entire OVC domain. We could further expand our model to incorporate other medical information such as examination results and treatment plans so that complete medical records can be computerized. Last but not the least, we could add a decision support module so as to actively assist doctors in making better judgments and decisions.


  1. E. Nygren and P. Henriksson, 1992. Reading the medical record I, Computer Methods and Programs in Biomedicine, 39: 1 - 12.

  2. B. Shneiderman, 1992. Designing the User Interface: Strategies for Effective Human-Computer Interaction, 2nd Edition, Reading, Mass.:Addison-Wesley.

  3. H. K. Walker, 1973. The problem-oriented medical record: an introduction, (in) Applying the Problem-Oriented System, ed., H. K. Walker, New York: Medcom Press Incorporated.