Multiple Criteria Approaches for Medical Decision Support Models

Abstract

It is not easy to decide on a final solution among a set of alternatives considering multiple criteria; therefore, several Multiple Criteria Decision Making (MCDM) approaches have been proposed and implemented in the literature. When uncertain data is involved in the decision process, the task gets even more difficult. In this thesis, we propose three different approaches to present evaluations of alternatives to decision makers (DMs) in such situations. All approaches focus on outranking relations of alternatives, and they utilize the well-known method Preference Ranking Organization Method for Enrichment Evaluation (PROMETHEE) in developing their evaluation measures. However, PROMETHEE rules are modified for uncertain data, and several statistical and probabilistic analyses are used to reach comprehensive outputs for decision support. The uncertain data can be obtained from previous observations, expert evaluations or samples from a probabilistic model such as Bayesian Networks (BNs). Two of the proposed approaches are the test-based and score-based approaches; they provide different levels of flexibility to the DM by offering partial and complete ranking of alternatives, respectively. The third approach, probabilistic PROMETHEE, offers both partial and complete alternative rankings using joint probability distributions of alternative evaluations in each criterion. Two different case studies are conducted to test the approaches: medical treatment selection for shoulder pain is studied to test all three approaches and also, a supplier selection case is studied to assess all approaches in a different domain. Additionally, sensitivity analyses are performed to test the sensitivity of alternative rankings to changes in criteria weights.