Change search
ReferencesLink to record
Permanent link

Direct link
Automated Annotation of Events Related to Central Venous Catheterization in Norwegian Clinical Notes
Norwegian University of Science and Technology, Faculty of Information Technology, Mathematics and Electrical Engineering, Department of Computer and Information Science.
2014 (English)MasteroppgaveStudent thesis
Abstract [en]

Health personnel are required to use Electronic Health records for documentation and commu- nication. Clinical notes from such records contain valuable information, but unfortunately this is often in narrative form, making it difficult to retrieve and extract information from them. One such problem is to get an overview of the number of patient days for patients with central venous catheter (CVC). The risk of infections increase with an increasing number of patient days. The present study examines the utility of applying NER to extract CVC related events from clinical notes. No studies have previously examined this application for Norwegian Clinical notes. Conditional random fields are used to make models based on different feature sets. The feature sets are combinations of word window, stem, synonymous and International classification for Nursing Practice (ICNP) axis. A corpus manually annotated with CVC event types was used for training and testing different models using three-fold cross-validation. Sixteen different combinations of features were tested. A factorial analysis using the three cross-fold runs as blocks was conducted to determine which features had the greatest effect on performance. Word window, ICNP axis and an interaction effect between these were found to affect performance significantly. Stem had an effect on recall, whereas no such effect was found for precision. An interaction effect between synonymous and ICNP-axis was found to effect precision. Accumulative scores of the different label types gave a precision of 56.29 %, a recall of 39.4 % and a f-measure of 46.33 for the best feature combination. Overlapping labels, errors in corpus and manual annotation are sources of error in the study. Thus, further research is necessary to draw certain conclusions about the present findings.

Place, publisher, year, edition, pages
Institutt for datateknikk og informasjonsvitenskap , 2014. , 84 p.
URN: urn:nbn:no:ntnu:diva-25068Local ID: ntnudaim:8408OAI: diva2:729999
Available from: 2014-06-26 Created: 2014-06-26 Last updated: 2014-06-26Bibliographically approved

Open Access in DiVA

fulltext(12261 kB)442 downloads
File information
File name FULLTEXT01.pdfFile size 12261 kBChecksum SHA-512
Type fulltextMimetype application/pdf
cover(184 kB)0 downloads
File information
File name COVER01.pdfFile size 184 kBChecksum SHA-512
Type coverMimetype application/pdf

By organisation
Department of Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 442 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 246 hits
ReferencesLink to record
Permanent link

Direct link