Decision-theoretic planning of clinical patient management

Abstract

When a doctor is treating a patient, he is constantly facing decisions. From the externally visible signs and symptoms he must establish a hypothesis of what might be wrong with the patient; then he must decide whether additional diagnostic procedures are required to verify this hypothesis, whether therapeutic action is necessary, and which post-therapeutic trajectory is to be followed. All these bedside decisions are related to each other, and the whole task of clinical patient management can therefore be regarded as a form a planning. In Artificial Intelligence, planning is traditionally studied for situations that are highly predictable. An important characteristic of medical decisions is however that they often must be made under conditions of uncertainty; this is due to errors in the results of diagnostic tests, limitations in medical knowledge, and unpredictability of the future course of disease. Decision making under uncertainty is traditionally studied in the field decision theory; in this thesis, we investigate the problem of clinical patient management as action planning using decision-theoretic principles, or decision-theoretic planning for short.