The point of view discussed here is that of a clinician in the Medicine Department of the Geneva University Hospital. This is a non computer scientist approach of a generalised representation of medical knowledge. Two important issues are discussed in this paper. The first issue is concerned with defining the necessary contents of what should be modelled. An attempt is made to define domains that should be included in the model. The extent of the model may be defined in two ways. Horizontally, by meaning the various domains to be included such as anatomy, therapy, and vertically to define the granularity or the precision with which each domain will be defined. The knowledge included in such a model may cover natural language processing, expert and alert systems, coding schemes and bibliographic retrieval.
The second issue discussed in this paper is how to model medical knowledge. One of the features of the medical knowledge is its relative evolution, which needs good maintenance tools. The validation and coherence of such a model will be difficult to maintain, depending on the experts in charge. Furthermore, such a model will probably need to be built by medical people rather than by computer scientists. This requires a user-friendly interface for entering new data and for control tools. The system must include a knowledge browser, a coherency and redundancy checker and tools to put together different models built on the same platform. In order to stand any chance of imposing itself on the various actual coding schemes, such a system must be language and hardware independent. As a concrete example, the Geneva experience with a modelling system called the Galen Master Notation, developed in the context of the European project Galen, is described.