Structural, elicitation and computational issues faced when solving complex decision making problems with influence diagrams

Abstract Influence diagrams have become a popular tool for representing and solving decision making problems under uncertainty (Shachter, Operations Research 1986;34:871–82). We show here some practical difficulties when using them to construct a medical decision support system. Specifically, it is hard to tackle issues related to the problem structuring, like the existence of constraints on the sequence of decisions, and the time evolution modeling; related to the knowledge-acquisition, like probability and utility assignment; and related to computational limitations, in memory storage and evaluation phases, as well as the explanation of results. We have recently developed a complex decision support system for neonatal jaundice management — a very common medical problem — , encountering all these difficulties. In this paper, we describe them and how they have been undertaken, providing insights into the community involved in the design and solution of decision models by means of influence diagrams. Scope and purpose Decision Analysis is a very well-known discipline that deals with the practice of Decision Theory (Clemen, Making hard decisions: an introduction to decision analysis, 2nd ed. Pacific Grove, CA: Duxbury, 1996). It comprises various steps usually implemented in a decision support system: definition of the alternatives and objectives, modelization of the structure of the decision problem, as well as the beliefs and preferences of the decision maker. The recommended alternative is the one with maximum expected utility, once all the assignments have been refined via sensitivity analyses. However, there are a number of difficulties faced in practice when solving large problems, that require an attentive study.