Rough Classification of Pneumonia Patients using a Clinical Database

Abstract

This study used the original model of rough sets [1] for data analysis of objective clinical findings from pneumonia patients. Pawlak’s rough classification algorithm [2] was used to find the reduct, which is a logical construct of the most information-preserving findings from a decision table. The condition attributes were the clinical findings that were used by a hospital information system, MedisGroups, as independent variables in the disease severity scoring algorithm for Bacterial Lung Infection or Other Lung Infection diseases. The International Classification of Diseases (ICD) code on the patient’s medical record was used as the decision attribute. The condition attributes not included in the reduct are considered superfluous with respect to the decision attribute. Six of the twenty-five condition attributes formed the reduct.

Some diseases, such as pneumonia, do not have a gold standard for validating a diagnosis. Iliad, an expert system based on Bayes’ Theorem, was chosen for evaluation of the rough classification results. The same subset of condition attributes appeared in both the rough sets logical classifier and Iliad’s probabilistic classifier.

In addition, a machine learning system, LERS (Learning from Examples based on Rough Sets), was used to induce rules from the decision table.