Classic versus sequential diagnostic support for chronic nonspecific respiratory diseases.

Abstract

The authors discuss two approaches to assisted medical diagnosis of chronic nonspecific respiratory diseases, Bayes-Fisher linear discrimination and a decision tree based on the linear discriminant functions of the individual nodes. Only the two most discriminative variables obtained for differentiation of groups at individual nodes are used for the node subclassification. They are the four linguistic variables cough, character of dyspnea, physical examination of the chest, and x-ray examination of the chest obtained by conversion of discrete features. No laboratory finding is used. The tree classifier simulates a physician's thought process. Results with multigroup linear discrimination on the same four variables for both a reclassification and a cross-validation technique were comparable to the tree-scheme outcomes. The tree classifier lends itself to simple graphic presentation, the patient being created as a point in the plane of the two most discriminative variables without the need for mathematical formulas and posterior probabilities. The classification errors that appear in the chronic bronchitis group can be easily corrected using an additional variable, smoking index.