OBJECTIVES: The goal of this research was to design a solution to detect non-reported incidents of injection.

METHODS: We developed methods to process electronic medical records and automatically extract clinical notes describing incidents of injection by using the SVM based technique. First, we manually labeled a training set of clinical notes into two categories based on whether they included an incident report of injection or not, and then the machine learning models were created. This machine learning method arranges data in a vector space, using single words as the axes. For the training process, a few variations were tested: normalized versus non-normalized data.

The extracted notes are treated as incident candidates which are shown to the safety management department for further analysis.

RESULTS: Using the developed method based on the SVM, we implemented an incident candidate reporting system. To evaluate the system, we asked a staff of the safety management department to judge whether extracted incident candidates were incidents or not. The system used inpatients’ clinical notes written from January 8, 2018 to January 14, 2018 in Kyoto University Hospital.

As a result, in the case of non-normalized data, 41 out of 364 incident candidates were judged clinical notes describing incidents. Furthermore, 21 of them were non-reported incidents. In the case of normalized data, 23 out of 91 incident candidates were judged clinical notes describing incidents. Moreover, 13 of them were non-reported incidents.

CONCLUSIONS: In this research, we aimed to establish a method to extract incident candidates from clinical notes in order to detect non-reported incidents of injection. In addition, we created a reporting system that presents incident candidates extracted by using the developed method. The system successfully detected non-reported incidents to the safety management department, thus our goal was achieved.