革新知能統合研究センター

Title : Longitudinal Analysis of Medication History using Frequent Sequence Pattern

Abstract: Electronic Medical Records (EMR) has been accumulating a big data which harbors insight and knowledge for a retrospective study towards the development of evidence-based medical guidelines. And frequent sequential pattern (FSP) mining is proven to be a practical tool for analyzing such a huge amount of data. Previous methods based their algorithm on Apriori. However, these methods consist of properties that non consecutive sequences and partial itemsets are able to support the frequent patterns. Such properties generate a large number of frequent patterns that inhibit domain expert to analyze the interesting patterns. We present a strategy to do a longitudinal analysis upon medication history of chronic patients using frequent sequence pattern and some early results. The strategy developed so far consists of preprocessing and mining methods from which we are able to compress the search space and to produce a more compact resultset compared to the conventional method.