Decision support for diagnosis of lyme disease.

This paper describes the development of a Bayesian model for diagnosis of patients suspected of Lyme disease, and the integration of such a model into a medical information system. A Bayesian network incorporating the clinical history and laboratory results has been constructed. Because many of the symptoms are not exclusive to Lyme disease and they develop over time, the clinical history is important for making the correct diagnosis. The model is based on time slices, where each time slice contains the observed pathological picture from one consultation with for example, the general practitioner. Since the time intervals between consultations typically are not equivalent, we have developed a novel method that can handle non-equivalent time intervals between the time slices in the network. The method is based on a description of the general development pattern of Lyme disease, which is implemented in a model that states the conditional probabilities of experiencing a certain pathological picture given time since infection. The model has been integrated into a web-based medical information system, called Borrelia Systems, which has enabled us to evaluate the model during a progressive diagnostic process. The integration has been accomplished through the development of a Bayesian Application Framework. This framework specifies a communication data structure in XML providing a graphical user interface and database components, which can be used when developing systems that are based on Bayesian networks. The framework generalizes the integration of Bayesian networks so that it is possible to switch network without manually having to update or change the system.