| One of the most frequent decisions a physician makes is whether to order a test, and the selection of a proper test or a battery of tests is often essential to the making of correct diagnoses and selection of the most appropriate treatments for a patient.;The selection of a battery of tests is complicated because there are usually a large number of alternatives for: test combinations, rules for interpreting compound results, and test performance sequences. In this research, test selection for binary, as well as multiple category, classification problems are addressed systematically in a multiobjective optimization framework.;Test selection problems are further complicated due to the mutual impacts among the tests or treatments selected in different stages of a patient-care process. In this research, the method of multiobjective multistage impact analysis (Gomide and Haimes, 1984) is combined with the envelope approach of the multiobjective dynamic programming (Li and Haimes, 1985) to develop a new method for solving multistage test/treatment selection problems.;For most part of this research, information regarding the test dependences, e.g., the joint result distribution of tests, is assumed to be unavailable, therefore, most of the models are based on certain subjective assumptions about the test dependence. A sensitivity analysis is then conducted to measure the impact that the potential levels of test dependence may have upon the calculated performances of a battery of tests. Based on that analysis, a more realistic assumption about the level of test dependence is suggested. Finally, considering that the joint result distribution for a group of tests is known, a linear programming model is then developed and used to find the best decision rule for a battery of tests.;The models developed in this research are demonstrated on simulated and real data bases. |
