Knowledge Base for Evidence Based Medicine with Bioinformatics Components

This paper presents an approach for a multilevel knowledge base system for evidence-based medicine. A sequence of events called patient trial is extracted from computer patient records. These events describe one flow of therapy for a concrete disease. Each event is represented by state and time. We introduce a measure between states, which is used to calculate the best alignment between different patient trials. The alignment measure calculates the distance between two sequences of patient states, which represents the similarity of the course of disease. Based on that similarityvalue classes are introduced by using specific clustering methods. These classes can be extended by gene expression data on micro-arrays leading to finer clustering containing similar trials called trial families. For easy checking if a new trial belongs to a family we use profiles of Hidden Markov models to detect potential membership in a family.