Planning of therapy and tests in causal probabilistic networks

摘要

Causal Probabilistic Networks (CPNs) provide a common framework that unifies the various tasks involved in medical reasoning. These tasks include causal and diagnostic reasoning, simulations, planning of tests and therapies and automatic updating of knowledge. While most of these tasks are quite well understood and have already been described in the literature, planning of therapy and tests is still a relatively unexplored area. In this paper we suggest planning methods that can be directly integrated in the CPN formalism. The methods are based on decision theory, and are realized by associating utilities with selected nodes in the CPN. The methods are simpler than planning methods based on influence diagrams and can be applied to planning problems of non-trivial size. This is illustrated by applying the proposed formalism to planning of insulin therapy in insulin-dependent diabetic patients. Planning of tests is illustrated by a small test/treat example.