Early detection of artherosclerotic coronary heart disease (CHD) is an important and difficult medical problem. CHD risk factors include artherosclerotic attributes, living habits, hemostatic factors, blood pressure, and metabolic factors (Goldman et al., 1996). Their screening is performed in general practice by data collection in three different stages.
A database with 238 patients representing typical medical practice in CHD diagnosis, collected at the Institute for Cardiovascular Prevention and Rehabilitation, Zagreb, Croatia, was used for subgroup discovery. The database is in no respect a good epidemiological CHD database reflecting actual CHD occurrence in a general population, since about 50% of gathered patient records represent CHD patients. Nevertheless, the database is very valuable since it includes records of different types of the disease. Moreover, the included negative cases (patients who do not have CHD) are not randomly selected persons but individuals with some subjective problems or those considered by general practitioners as potential CHD patients, and hence sent for further investigations to the Institute. This biased data set is appropriate for CHD risk group discovery, but it is inappropriate for measuring the success of CHD risk detection and for subgroup performance estimation in general medical practice.