Mining Association Rules from Clinical Narratives

We propose a method that processes raw informal medical texts (from health forums) and formal texts (outpatient records) in Bulgarian language in order to extract typical word co-occurrences in the form of association rules. When mining these rules we use some context information and small terminological lexicons to generalize the extracted frequent patterns. This allows to study informal expressions of medical terminology and to identify automatically typical descriptions of types of patient statuses. The paper presents association rules generated from 300,000 outpatient records and 1,425 forum postings and discusses their evaluation and usefulness. Employing this unsupervized data mining approach we hope to overcome the lack of linguistic resources that can support successful NLP analysis of clinical narratives in Bulgarian.