Summary
The aim of this study was to develop a computer expert system that could reproduce a pathologist's diagnosis of iron deficiency from the data obtained from blood tests. 275 cases were collected for construction and testing of the expert system. The expert system used a combination of fuzzy set logic and cut-off points from 14 parameters to arrive at one of 5 diagnostic categories graded from “iron deficient” to “no evidence of iron deficiency”. The agreement between pathologist and expert system was 0.91 (Spearman rank correlation coefficient) in the learning population; this dropped to 0.79 in the test population. Absolute agreement on diagnostic category was reached in 71% of cases. In no case was there disagreement by more than 3 grades.