Analysis of Medical Documents with Text Mining and Association Rule Mining

Abstract

Text mining techniques extracts meaningful information from large amounts of semi-structured and unstructured texts. In this work, the MetaMap tool was used to extract medical entities like diseases and syndromes from discharge summaries. Also, association rule mining algorithms such as Apriori and FP-Growth were applied to the extracted entities in order to find associations between them. The dataset used consists of 1237 discharge summaries obtained from the 2008 i2b2 Obesity Challenge. The rules that have a principal diagnosis as antecedent showed that the cardiac disease frequently occurred with other diseases like hypertension and diabetes. Most of the rules describe associations between diabetes and other diseases like hypertension, dyslipidemia, nephropathy, heart disease, lung diseases, and arthritis. These rules have a confidence parameter of above 0.5.