SOM: feature extraction from patient discharge summaries

Authors:
D. J. Tufts-Conrad;A. N. Zincir-Heywood;D. Zitner
Affiliations:
Dalhousie University, Halifax, NS, Canada;Dalhousie University, Halifax, NS, Canada;Dalhousie University, Halifax, NS, Canada
Venue:
Proceedings of the 2003 ACM symposium on Applied computing
Year:
2003

Citing 5
Cited 0

Concept features in Re:Agent, an intelligent Email agent

AGENTS '98 Proceedings of the second international conference on Autonomous agents
Machine learning in automated text categorization

ACM Computing Surveys (CSUR)
Neural Networks: A Comprehensive Foundation

Neural Networks: A Comprehensive Foundation
Exploration of Document Collections with Self-Organizing Maps: A Novel Approach to Similarity Representation

PKDD '97 Proceedings of the First European Symposium on Principles of Data Mining and Knowledge Discovery
Self organization of a massive document collection

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

In each Canadian province, hospitals collect information, at discharge, on the hospital stay of each patient. The information is collected in the form of a patient discharge abstract (PDA) and sent to the Canadian Institute for Health Information. The Patient Discharge Abstract uses the ICD-10-CA code standard to outline the assigned diagnoses for the patient's condition and the procedures that were performed. One compulsory piece of information in the Patient Discharge Abstract is the identification of the "most responsible diagnosis" (MRDx) -- that diagnosis considered to be the most significant condition of the patient that caused the greatest length of stay in hospital. This research investigates the potential for automating the process of feature extraction from a narrative patient discharge summary to support the classification of the MRDx for a Patient Discharge Abstract. Unsupervised neural networks -- Self-Organizing Maps (SOM) -- are effective for classification tasks based on noisy input patterns. Here a hierarchical architecture of SOMs is used to identify semantic similarities encoded in the original information and visualize the characteristics of an MRDx.