You are here
Issues in the Classification of Disease Instances with Ontologies
Ontologies define classes of entities and their interrelations. They are used to organizedata according to a theory of the domain. Towards that end, ontologies provide classdefinitions (i.e., the necessary and sufficient conditions for defining class membership). Inmedical ontologies, it is often difficult to establish such definitions for diseases. We usethree examples (anemia, leukemia and schizophrenia) to illustrate the limitations ofontologies as classification resources. We show that eligibility criteria are often moreuseful than the Aristotelian definitions traditionally used in ontologies. Examples ofeligibility criteria for diseases include complex predicates such as 'x is an instance of theclass C when at least n criteria among m are verified' and 'symptoms must last at least onemonth if not treated, but less than one month, if effectively treated'. References to normalityand abnormality are often found in disease definitions, but the operational definition ofthese references (i.e., the statistical and contextual information necessary to define them) israrely provided. We conclude that knowledge bases that include probabilistic andstatistical knowledge as well as rule-based criteria are more useful than Aristoteliandefinitions for representing the predicates defined by necessary and sufficient conditions.Rich knowledge bases are needed to clarify the relations between individuals and classes invarious studies and applications. However, as ontologies represent relations amongclasses, they can play a supporting role in disease classification services built primarily onknowledge bases.