Obstacles to computerised manipulation and intuitive user interfaces
Simple classifications contain little formal structure which can be manipulated by a computer. In traditional classifications, the only structure is the single hierarchy. The interpretation of the meaning of codes is largely in their rubrics for interpretation by human users rather than in the model itself where it is accessible to automatic manipulation. Some systems provide additional structure in synonyms, quasi-synonyms (the dagger-asterisk links of ICD), but the meaning of these links is also variable and requires human interpretation.
As a result, the amount of manipulation which is possible, whether for decision support, user interface, or information retrieval is limited. Human use of concept systems is so natural that it is easy to underestimate the subtlety and complexities involved. User interfaces of systems using simple hierarchical systems have shown serious limitations. If computer systems are to manipulate concepts in ways which seem intuitive to human users, then much more of the structure of those concept systems must be modelled formally so as to be accessible to the computer.
In order to make the point more clearly, consider the knowledge which needs to be manipulated to deal sensibly with the concept of Pathological fracture of the shaft of the femur caused by osteoporosis. Clearly this concept is made up of a number of different elementary concepts. Potentially it might be classified as being a kind of the concept of fractures of femur, the concept of pathological fractures, the concept of conditions caused by osteoporosis, or even fractures of the shaft of long bones. It might even be considered whether pathological is redundant, since a fracture caused by osteoporosis might be considered to be pathological by definition.
What knowledge needs to be represented in order to produce a model of our understanding of this concept? A short list must include:
- Fractures exist and are conditions
- Fractures can be pathological
- Femurs exist and are bones
- Fractures may occur to the femur or more generally to bones.
- The femur has a part (division) known as the shaft
- The shaft of the femur may fracture or more generally, divisions of bones may fracture
- Fractures of the shaft of the femur are fractures of the femur or more generally, fractures of divisions of bones are fractures of the corresponding bones, or more generally still that conditions of divisions of things are conditions of the corresponding thing.
- What do we mean by pathological in pathological fractures? Are pathological fractures defined as: a) fractures in which there is a cause other than external trauma? b) fractures in which external trauma is definitely not one of the causes? or c) fractures for which the role of external trauma is not specified? d) In some other way?
To capture all of this information within a simple hierarchical classification system is implausible. A richer structure is needed.
Note that we are not saying that everything about the concepts involved need be captured. There is much knowledge about the concepts involved which we do not need to represent in order to determine how to combine entities and classify them correctly. For example we do not need to know the details of the anatomy of the femur or the detailed relation of the shaft of the femur to the femur itself. Nor do we need to know anything about osteoporosis other than that it is a condition which can cause fractures. That osteoporosis involves decalcification is very interesting, might even be needed elsewhere in the model for classifying osteoporosis, but is unnecessary here. Attempting to represent too much information has frequently been as serious a practical barrier to computer manipulation as representing too little. The task proposed must be practical for the human developers and computationally tractable.