Test-selection Strategies for Probabilistic Networks

Abstract

Decision-support systems are used in a large variety of domains. In the medical domain, such systems can be equipped with a patient-specific facility that indicates which diagnostic test or combination of tests should be performed. Current systems, however, do not take into account that tests might be ordered in packages rather than one by one. Current systems furthermore do not take into account that physicians would prefer to gather information for a specific purpose, such as the general condition of the patient, prior to gathering information about an other topic, such as for instance the presence or absence of metastases. In this thesis, we describe a automatic test-selection facility we designed that fits better with the daily medical practice than current facilities do. Our facility is capable of selecting diagnostic tests while taking into account the specific characteristics of the patients illness, the characteristics of the test, such as its sensitivity and specificity or its predictive value, and the order in which various different subgoals are investigated by the physicians. It is furthermore able to select multiple tests rather than just one test. By extending a decision-support system with a test-selection facility, the system become more and more interesting to use in the daily medical practice. A decision-support system with a good test-selection facility could lead to ordering fewer tests. This might imply lower financial costs, shorter periods of waiting prior be able to visit the physician or receive therapy. Futhermore, it might lead to better patient care and a higher quality of care in general.