A Goal-Based Framework for Supporting Medical Assistance: The Case of Chronic Diseases

Abstract

Chronic diseases cause many deaths worldwide each year. The most common action to manage this type of disease is to conduct expensive pathological tests whose goal is to assess risks and identify early complications to patients health or to prevent the patient from developing other diseases. The choice of which action/test to perform is therefore important. There are in the literature some guideline-based decision-support systems which help to assist practitioners to choose the appropriate therapeutic action for their patients. The idea of such systems is to structure a guideline as a set of choices to be made by the clinician. However, such a set of actions/tests/decisions can change according to new information obtained concerning the patient state (direct answers from the patient, blood test results, etc.). Because the time factor may be crucial and because executing tests is expensive, it is then important for the practitioners to ask for or to quickly obtain useful information helping take the good decision in order to achieve the goals of excluding the risks as soon as possible, which means with a minor quantity of information/tests required. In this paper, we address the challenges introduced above by proposing an agent-based framework that supports the development of an intelligent goal-driven agent to help practitioners in choosing the most useful action to perform (question to ask a patient, test, etc.) in the case of a chronic disease. The framework supports the selection of the next dialogue action by measuring the usefulness, with respect to a goal, of a piece of information to be obtained. We introduce our framework through the use of a running example, showing how an agent can drive the interaction based on both its background knowledge and the new information it acquires. Experiments performed concerning two chronic diseases, namely asthma and type-2 diabetes, validate our approach. Finally, we discuss further possible scenarios where our framework can be applied in different ways.