A CLINICAL DECISION METAKNOWLEDGE MODEL Joël COLLOC , Marie-Hélène CUVELIER CRIS-LEACM, Université de LYON Université Lyon2, ISH, 14-16 Avenue Berthelot 69007 Lyon, France Joel.colloc@ish-lyon.cnrs.fr mariehelenecuvelier@wanadoo.fr Peter SUMMONS The University of Newcastle University Drive, Callaghan, NSW 2308, Australia Peter.Summons@newcastle.edu.au ABSTRACT: This paper presents a clinical metaknowledge model that can be applied in a wide range of application domains that use Knowledge Base Systems (KBS). It relies on a general description of the clinical decision process. This model can be then specialized to solve similar problems in all domains where processes similar to clinical decisions are applied. We show the relevance of our model in three different domains: medicine; management; and education. KEYWORDS: metacognition, management, medicine, education, prognosis, living being, case based reasoning, multi-agent systems. INTRODUCTION We propose a clinical metaknowledge model to build Decision Support Systems (DSS) and Knowledge Bases (KB) in different scientific and technical areas where the clinical approach is relevant. Section 1 presents the model that provides us with a framework to express the clinical decision process independent to any specific application domain. The model relies on a definition of the main steps of the clinical process and provides operators to specialize and adapt these steps to other situations and domains. Section 2 presents the clinical decision process used in medicine. Section 3 shows how the same process can be applied in management and explains why a company can be considered as a living being. Section 4 looks at the education domain, where one can find similar clinical steps used to cope with the cognitive deficiencies of some pupils. Finally, Section 5 describes methods to implement the metaknowledge model. 1. A METAKNOWLEDGE CLINICAL MODEL We propose a metaknowledge model, which relies on the decision process similarities found between practitioners’ behaviour with other specialists and can be seen in many varied domains such as medicine, management and education. A decision process problem can be defined as a choice between several options in order to achieve a goal as efficiently as possible. Four main steps define the model: diagnosis; prognosis; treatment; and follow up. These steps make up the clinical decision loop. Each step is a complex task that involves similar means and moreover similar goals. These main goals are: to recognise a situation as a clinical picture; to make a prognosis of the situation, firstly if nothing is done, and secondly if something is done to modify the current evolution. Thus, the prognosis should predict the probable outcome of the situation. According to the prognosis, treatment goals are established that try to enhance the situation and, if possible, to solve the problem. Most of the time, it is necessary to monitor and control the treatment’s efficiency and the possible occurrence of side effects that might cause new problems (that would then need to be diagnosed). The complete clinical decision loop is illustrated in Figure 1. 2. MEDICINE 2.1. the medical context The medical decision activity could be defined as a decision chain (more exactly a network) that involves the following steps/stages: diagnosis; prognosis; therapeutic and treatment follow-up. As some of these steps become more and more complex, a decision aid tool would be useful, especially during the prognosis and therapy stages. Paradoxically, many physicians are interested in information systems and some of them have developed small, or sometimes more sophisticated, Decision Support Systems (DSS). However, these systems are very locally used inside a hospital or a research department by a small group of specialists who have to solve very specific and localised problems. Identify a known situation or symptoms Diagnosis Δ Prognosis Π Treatment Θ Follow-up Φ Conflict, Crisis, Dysfunction… Action goals Side effects A new problem ? Figure 1. a clinical metaknowledge model Stable state Σ routine control Lethal state Ω No more treatment