This project aimed to automatically construct transition probabilities from clinical data; these probabilities are used for dynamic decision modeling. Dynamic decision models are frameworks for choosing alternative courses of action when the consequences of such choices are uncertain. Frequently, the numeric parameters of dynamic decision models, such as transition probabilities, are subjectively determined by medical experts or painstakingly derived from literature. However, when the decision situations are complex or the decision dimensions are large, the practicality of the modeling approach is limited by the lack of realistic estimations. On the other hand, calculating objective probabilities from clinical data may be complicated by recording formats, measurement assumptions, and processing errors. We examined some of these issues by formulating a simplified dynamic decision model in therapy management for diabetic patients.