A Bayesian expert system for clinical detecting coronary artery disease

Background: The purpose of this study was to use a data mining technique to develop an expert system of the Bayesian model for detecting coronary artery disease (CAD). In addition, this study provides an evaluation of CAD detection before an invasive cardiac angiography as well as a paradigm for implementing relevant expert systems in the future. Methods: The study samples were drawn from all patients with cardiac angiography between August 1, 2005 and July 31, 2006, from the cardiac department in a medical center (Tri-Service General Hospital, TSGH), excluding samples with acute myocardial infarction, dilated cardio myopathy and rheumatic heart disease. A total of 415 samples were studied. All CAD-related risk factors were data-mined using a training set of randomly extracted 204 samples. All risk factors were calculated for sensitivity and specificity for Bayesian modeling and the implementation of the localized rules of a knowledge based. Furthermore, this study also quoted the epidemiological results of the knowledge based external rules from the PROspective Cardiovascular Münster study (PROCAM). Two knowledge bases, the TSGH base and the PROCAM base, were validated by a testing set of 211 samples. Results: The accuracy rates of the TSGH and PROCAM bases were as high as 70%. For detecting CAD, the localized data mining of the TSGH-based AUC was more stable at 86.2%, outperforming the PROCAM-based AUC of 82.2%. Conclusions: In this study, an evidence-based clinical expert system of the Bayesian model provides an evaluation for detecting CAD before an invasive cardiac angiography as well as a paradigm for relevant expert systems.