Validation of a SPHINX knowledge base: some supplementary elements

Abstract

This study has been produced from one hundred medical records of various origins (Manuel-Michel, 1985). They were referred to five adjudicators concerned with diabetes therapy: two experts in the domain and three general practitioners who volunteered to take part in the study. Added to these five advisers were a doctor in attendance who had completed the record and given a prescription, and SPHINX. It therefore concerns a study of consensus involving seven hundred prescriptions. As in the evaluation of the MYCIN system, the study is based on a rigorous methodology dependent on a protocol which limits evaluation bias as much as possible. The initial objective was to place SPHINX among practitioners with different levels of expertise; and the second objective was the estimation of the degree of expertise of the system through a discerning study on the consensus and disagreement between the system and the evaluators.