gms | German Medical Science

Introduction: In medical practice, therapies sometimes do not give desired results. There are many different reasons. A diagnosis might be erroneous, the state of a patient might have changed, a patient might have caught an additional disease, some other complication might have occurred, or a patient might have changed his lifestyle (e.g. started a diet) and so on.

For long-term therapy support in the endocrine domain and in psychiatry, we have developed a Case-Based Reasoning system that not only performs typical therapeutic tasks but also especially deals with situations where therapies become ineffective. Therefore, it first attempts to find causes for inefficacy and subsequently computes new therapy recommendations that should perform better than those administered before.

Materials and Methods: The system is designed to solve typical problems, especially inefficacy of prescribed therapies that can arise in different medical domains. Therefore most algorithms and functions are domain independent. Another goal is to cope with situations where important patient data is missing and/or where theoretical domain knowledge is controversial.

The system does not generate solutions itself. Its task is to help users by providing all available information and to support them to find solutions. Users shall be doctors, maybe together with patients.

In addition to the typical Case-Based Reasoning knowledge, namely former already solved cases, further knowledge components are used, namely medical histories of the query patients themselves and prototypical cases (prototypes). Furthermore, the knowledge base consists of therapies, conflicts, instructions and so on.

The case history is written in the patient’s individual base as a sequence of records. A patient’s base contains his whole medical history, all medical information that is available: diseases, complications, therapies, circumstances and so on. Each record describes an episode in a patient’s medical history. Episodes often characterise a specific problem. Since the case base is problem oriented, it contains just episodes and the same patient can be mentioned in the case base a few times. Information from the patient’s individual base can be useful for a current situation, because for patients with chronic diseases very similar problems often occur again.

The knowledge base contains information about problems and their solutions that are possible according to the domain theory. It has a tree structure that is organised according to keys and it consists of lists of diagnoses, corresponding therapies, conflicts, instructions, and medical problems (including solutions) that can arise from specific therapies. The knowledge base also contains links to guidelines and references to correspondent publications.

The case base is problem oriented. Thus a case in the case base is just a part of a patient’s history, namely an episode that describes a specific problem that usually has a solution too. So, the case base represents decisions of doctors (diagnosis, therapies) for specific problems, and their generalisations and their theoretical foundations. Every case solution has (usually two) generalisations, which are formulated by doctors. The first one is expressed in terms of the knowledge base and it is used as a keyword for searching in the knowledge base. The second generalisation of a solution is expressed in common words and it is mainly used for dialogues.

Former cases in the case base are indexed by keywords. Each case contains keywords that have been explicitly placed by an expert. For retrieval three main keys are used: a code of the problem, a diagnosis, and a therapy. Further keys can be used optionally.

Prototypes (prototypical cases with generalized solutions) help to select a proper solution from the list of probable or available solutions. A prototype may help to point out a reason of inefficacy of a therapy or it may support the doctor’s choice of a drug.

Adaptation takes place as a dialogue between the doctor, the patient, and the system. The system presents different solutions, versions of them, and asks questions to manifest them. The doctor answers and selects suggestions.

Results: The system helps users to find solutions. Sometimes it just provides ideas. Since in the endocrine domain and in psychiatry it is very important what patients tell their doctors, the results depend on the whole dialog between the system, a doctor and the patient. So, an evaluation about how many of the systems results are right or wrong does not make any sense here. Instead, we have developed a questionnaire, which was filled in by three users (one in the endocrine domain, two psychatrists). The questionnaire contained 15 questions for which the users could give assessments from 1 (very poor) to 5 (very good). The outcome is not surprising. In the endocrine domain a strong theory exists and the doctor claims that he is very experienced, that he does not really need the system and that he uses it just to question his own decisions. So, he assessed the usefulness as rather moderate but he seems to be very satisfied with the quality of its recommendations. In psychiatry the situation is the other way round, the theory is rather weak. This can be the reason that the two doctors assessed the usefulness rather high.

Discussion: We have developed a CBR system that helps doctors to solve medical problems, particularly to investigate causes of inefficacy of therapies. It deals with two application areas. In the endocrine domain a strong theory exists and therefore usually a lot of solutions are theoretically possible. So, here the main task of the system is to reduce the set of possible solutions as much as possible.

In the psychiatry domain the theory is weak, although big theories exist, which partly contradict each other. Since single patients are too specific here, the main task is to search with the help of general solutions in the case base for better solutions.