| Summary: | 碩士 === 國立陽明大學 === 衛生資訊與決策研究所 === 94 === Background：In Taiwan ,the Department of Health built the first national patient safety reporting system in 2004. For saving the time and human resources to re-classify the reporting events. The aim of this study will find out an automatic classification system for patient safety reports.
Material and Method：The training set include 1265 reports from September to December in 2005.The experts in charge this system to classify reports and to sift keywords out from ten kinds of report types. The testing set include 575 reports from February to March in 2006.Using information retrieval method called “space vector model” to classify the testing report events, the experts reviewer classify the same set been the “Golden Standard”. We calculate Kappa, Recall and Precision between system and experts to measure the performance of this system.
Result：This program finds out 182 keywords in 10 types event. The Kappa value of the medicine was 0.82、the fall was 0.82、and the tube was 0.86,the performance were almost prefect. The whole Recall value was 0.56, in which the medicine was 0.90,the fall was 0.89 and the surgery was 0.82.The whole Precision value was 0.55, in which the medicine was 0.80、the fall was 0.86、the blood transfusion was 1、the security was 0.88 and the harm was 0.88. The performances of the other events were not ideal.
Conclusion：Using information retrieval technology will classify the patient safety reports automatically. We will search and collect more keywords in all type events to improvement the performance of the system.
|
|---|
