This chapter introduces a case based reasoning (CBR) system for customizing treatment processes. The CBR system enables the generating of inpatient and outpatient treatment processes and the supporting of e-services in heath care networks individually customized to the patients’ needs. According to the CBR paradigm, which solves problems based on past experience, the proposed system uses old treatment processes of similar former patients and modifies them for new patients. In general, CBR is an established and well suited artificial intelligence method to support medical decision making. However, CBR systems capable of planning treatment processes by adapting old treatment processes to fit new patients are rare. The aim of this system is to increase the treatment quality of the patient by providing physicians with valuable treatment propositions and to contribute to the development of medical CBR Systems by introducing procedures enabling the generating of new treatment processes by modifying former treatment processes.
In the age of a growing flood of information on medical data and knowledge, medical decision support systems are becoming more and more important. Various artificial intelligence methods have been implemented in numerous systems to support the medical reasoning process. Among these methods case based reasoning (CBR) was established to support medical decision making. CBR (Amodt & Placa, 1994) solves problems by using past experiences and general knowledge. Past experiences are saved in form of cases in the case base. For solving a new arising problem the CBR system searches for similar problems and attempts to adapt their solutions to fit the new problem. CBR is particularly suited for helping find solutions to medical problems (Heinisch et al., 1998, p. 1) as it resembles the physicians’ cognitive process of recalling former patients and reusing past experiences. Furthermore, the collection of patient records which represent a valuable knowledge resource can easily be integrated in a CBR system as a case base.
Medical CBR systems can be divided into different categories depending on their purpose-oriented properties and their functional properties. Purpose-oriented properties (Nilsson & Sollenborn, 2004, p. 179) describe the general aim which the system fulfills and allow the separation of medical CBR systems into diagnostic systems, classification systems, tutoring systems and planning systems. While diagnostic systems are intended to support the whole diagnostic process, classification systems focus on special diagnostic problems (e.g. image classification). Tutoring systems aim at teaching students medical knowledge based on patient records. Planning systems provide assistance by configuring medical processes (e.g. therapies) consisting of several steps. Considering the functional properties CBR systems can be classified into case-match-systems and case-adaptation-systems (Goos, 1996, p. 15). Case-match-systems only enable the retrievement of similar patient cases, whereas case-adaptation-systems also allow the adaptation of past cases to fit new patients. Most of the medical CBR systems developed in the last years are classification systems or diagnostic systems and they support case-matching only (Nilsson & Sollenborn 2004, p. 182). Moreover, the majority of these systems focus on inpatient treatment and specialize in one certain disease. Planning systems which realize the adaptation task, which support inpatient and outpatient treatment processes in health care networks and which implement disease independent algorithms are missing.
This chapter describes such a CBR system. The system proposes inpatient and outpatient treatment processes in health care networks based on the adaptation of treatment processes to new patients and can be applied to the treatment of various diseases. Besides the treatment steps to be fulfilled by the physicians, the generated propositions also contain e-services to satisfy the coordination and informational needs of the health care providers and the patients.
In order to clarify the functionality of this appraoch, all functions of the CBR systems are illustrated by example of heart failure treatment. Heart failure is a syndrome which is caused by cardiac disorder and which weakens the pumping capability of the heart supplying the tissue with blood and oxygen (Hoppe & Erdmann, 2004, p. 11). It is one of the most frequent and severest diseases of the industrialized countries with an approximate prevalence ranging from 0.3% to 2% and an estimated incidence of 0.1% to 0.5% (Cowie et al., 1997, p. 211). The medical treatment of the multitude of patients is important. However, in practice many patients do not obtain an adequate therapie compliant to actual guidelines (Hoppe & Erdmann, 2004, p. 15; Stödter, 2000, pp. 6, 18f). So the application of the Medical CBR system to the treatment of heart failure will hopefully contribute to the improvement of this situation.
Key Terms in this Chapter
Maintenance: Preservation and recoverage of the Case Based Reasoning System. The aim of the maintenance is to garantuee high quality and high efficiency of the Case Based Reasoning System throughout its whole life-cycle (Wilson, 2001, p. 1). The tasks of the maintenance are monitoring the status of the Case Based Reasoning System in order to recognize changes, which decrease the capability of the Case Based Reasoning System, and executing counteractive measures.
Medical Case Based Reasoning: Case based reasoning for supporting medical decision making. Medical Case Based Reasoning Systems can be applied to different medical purposes. They are often used in diagnosis and classification (Nilsson & Sollenborn 2004, p. 182). Further fields of application are tutoring and planning therapies.
Case Based Reasoning: Method of Artificial Intelligence. It enables the solving of problems based on experience (Goos, 1996, S. 13). The experience is saved in the form of cases in a case base. Each case consists of two part: a decription of the problem and a description of its solution. In order to solve a new problem, Case Based Reasoning searches for the case of the case base which is most similar to the new problem and reuses its solution.
Structural Adaptation: Adaptation approach of Case Based Reasoning. It allows adapting the structure of the solution of a case from the case base in order to fit the problem of the query. Depending on the differences of the problem of the query and the problem of the similar case from the case base, components of the solution of the similar case are removed and reordered or new components are added (Wilke & Bergmann 1998, p. 500).
Extented K-D-Tree: Approach to save and retrieve knowledge in an efficient way. The basic idea is to partiton the knowledge base into k dimensions using a tree consisting of nodes and edges (Weß, 1995, pp. 180ff). While the leaf nodes of the tree save the knowledge, the inner nodes and the edges contain testing conditions, which direct the search of the knowledge. In Case Based Reasoning it can be used to save and search for cases efficiently.
Similarity Measures: Two-digit, real-valued function on the problem space: Sim: PxP ? [0;1]. The value range of similarity measures is restricted to the interval between zero and one, whereas the value zero represents minimal similarity and the value one symbolizes maximal similarity. Similarity measures enable the calculation of similarity between the problem of the query and the problem of the case of the case base (Stahl 2003, S. 47).
Gradient Descent Method: Method for searching a local minimum of a function. In order to find a local minimum, the algorithm moves in the direction of the negative gradient (vector of the partial differentiations) of the function. In Case Based Reasoning it can be used to optimize the global similarity measure according to the preferences of the user (Stahl, 2003, pp. 105ff). For this purporse an error function must be defined by comparing the order of the cases preferred by the user and the order calculated by the global similarity measure. The gradient descent method searches the minimum of this error function.
Object-Oriented Representation: Format to represent knowledge. In comparision to Attribute-Value-Vectors it provides better posibilities of representing structural knowledge. In this type of representation knowledge is characterised by classes. A concrete piece of knowledge is composed of a set of instances of these classes.
Attribute-Value-Vectors: Format to represent knowledge. It is often used to represent case knowledge in Case Based Reasoning (Richter, 2003, p. 412). The knowledge is characterised by a certain number n of attributes A1,A2,...,An. One concrete piece of knowledge P is specified by the vector of the attribute values a1,a2,...,an from the ranges R1,R2,...,Rn: P = (a1,a2,...,an) ? R1xR2x...xRn.