Abstract

In the Netherlands the European Injury Data Base (IDB), Dutch Injury Surveillance System (DISS) administrated by VeiligheidNL, is used for monitoring injuries treated at EDs. Until a few years ago all variables in IDB/DISS (such as injury mechanism, product involved, type of injury and body part involved) were coded manually by the staff of the ED. To reduce the administrative burden on the EDs we developed a system for automatic text recognition. From narratives on what happened and from the diagnosis as recorded in open text fields in the Hospital Information Systems, coded information was derived. The objective of this project was to investigate if automatic text recognition software could replace manual coding in the hospitals. After assessment of several software tools, we chose IBM Modeler as our tool to develop automatic text recognition to code IDB/DISS variables based on narratives derived from the Hospital Information Systems. We taught the system from scratch how to code information on accidents and injuries. All possible words had to be classified into libraries and we had to teach the system how to interpret sentences. After the implementation of the system all records were processed by the software and manually checked and corrected if necessary. We recorded what corrections were made in order to analyse how the software performed and what improvements had to be made to the automatic text recognition software. Comparison of data from autotext analysis before and after manual corrections was performed. Preparing proper text analysis is time consuming but it is possible to use automatic text recognition software to code information on injuries. The work that has been done by the software makes coding injuries at VeiligheidNL easier. Analysis on the recorded corrections has not been finalized yet, results will be presented at the conference.