In recent years, data mining has started to receive increasing interest as a method of complementing domain specific expertise in various spheres of human activity. Apart from data specific issues, a key particularity of many real world problems, such as medical diagnosis, are the costs involved, the most important being the test and the misclassification costs. This paper evaluates ProICET, a new system built around the ICET algorithm. The system has been previously benchmarked on classical medical data sets. Here, we use a real medical dataset to test the current version of our system. The comparative analysis confirms that ProICET is the best at cost minimization out of several successful classifiers, while keeping a good accuracy rate.