A Bayesian approach to assist in the diagnosis of coronary heart disease

Abstract

The objectives of this thesis were to design a method for evaluation of the diagnostic potential of available indicators of coronary heart disease (CHD) and to present a systematics, quantitative procedure to aiding in its diagnosis. A sample space of patients was divided into two mutually exclusive groups, those with angiographic evidence of CHD, and those with no CHD. Active duty or retired military men between the ages of 30 and 67 years constituted the sample space. Tests and risk factors were available in the medical literature that a doctor could view as an indicator or contraindicator of CHD. A vector of these possible indicators was established and the diseased group was compared to the non-diseased group in an effort to evaluate the diagnostic potential of the indicators. This was done by discriminant analysis in conjunction with a Bayesian method of weighting the importance of tests results. The important indicators were then used to formulate a model for diagnosing CHD based on a Bayes' decision technique.