Extraction and Quantification of Pack-years and Classification of Smoker Information in Semi-structured Medical Records

Electronic medical records contain a wealth of information that is potentially invaluable to many interested parties. However, the fact that most of these documents are of semistructured nature and are comprised of fragmented English free text, region-specific templates and clinical sublanguage among many other things, has made it difficult to use existing Natural Language Processing tools on them directly and to extract those information. In this work, we focus our attention on a set of medical records pertaining to Rheumatoid Arthritis patients and we present a pattern-based methodology for extracting and quantifying pack-year information. We also introduce an extension to those patterns in classifying individual instances within these documents into a set of predefined smoker status classes. Since our effort in extracting pack-years from medical documents is the first in its kind to the best of our knowledge, we evaluate our approach on a manually selected document collection and present very promising results. We also evaluate our instance classification approach using an additional document collection. Appearing in Proceedings of the 28 th International Conference on Machine Learning, Bellevue, WA, USA, 2011. Copyright 2011 by the author(s)/owner(s).