Lexically-Triggered Hidden Markov Models for Clinical Document Coding

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 742–751, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Lexically-Triggered Hidden Markov Models for Clinical Document Coding Svetlana Kiritchenko Colin Cherry Institute for Information Technology National Research Council Canada {Svetlana.Kiritchenko,Colin.Cherry}@nrc-cnrc.gc.ca Abstract The automatic coding of clinical documents is an important task for today’s healthcare providers. Though it can be viewed as multi-label document classification, the cod- ing problem has the interesting property that most code assignments can be supported by a single phrase found in the input docu- ment. We propose a Lexically-Triggered Hid- den Markov Model (LT-HMM) that leverages these phrases to improve coding accuracy. The LT-HMM works in two stages: first, a lexical match is performed against a term dictionary to collect a set of candidate codes for a docu- ment. Next, a discriminative HMM selects the best subset of codes to assign to the document by tagging candidates as present or absent. By confirming codes proposed by a dictio- nary, the LT-HMM can share features across codes, enabling strong performance even on rare codes. In fact, we are able to recover codes that do not occur in the training set at all. Our approach achieves the best ever per- formance on the 2007 Medical NLP Challenge test set, with an F-measure of 89.84. 1 Introduction The clinical domain presents a number of interesting challenges for natural language processing. Con- ventionally, most clinical documentation, such as doctor’s notes, discharge summaries and referrals, are written in a free-text form. This narrative form is flexible, allowing healthcare professionals to ex- press any kind of concept or event, but it is not particularly suited for large-scale analysis, search, or decision support. Converting clinical narratives into a structured form would support essential activi- ties such as administrative reporting, quality control, biosurveillance and biomedical research (Meystre et al., 2008). One way of representing a docu- ment is to code the patient’s conditions and the per- formed procedures into a nomenclature of clinical codes. The International Classification of Diseases, 9th and 10th revisions, Clinical Modification (ICD- 9-CM, ICD-10-CM) are the official administrative coding schemes for healthcare organizations in sev- eral countries, including the US and Canada. Typi- cally, coding is performed by trained coding profes- sionals, but this process can be both costly and error- prone. Automated methods can speed-up the cod- ing process, improve the accuracy and consistency of internal documentation, and even result in higher reimbursement for the healthcare organization (Ben- son, 2006). Traditionally, statistical document coding is viewed as multi-class multi-label document classifi- cation, where each clinical free-text document is la- belled with one or several codes from a pre-defined, possibly very large set of codes (Patrick et al., 2007; Suominen et al., 2008). One classification model is learned for each code, and then all models are ap- plied in turn to a new document to determine which codes should be assigned to the document. The drawback of this approach is poor predictive perfor- mance on …

Content truncated to 500 words