Automatic Identification Of Security Incidents In Patient Complaints

[Purpose] With the widespreading attention to patient safety management in China,the study of patient safety has also attracted the attention of researchers.This study attempts to mine the information related to patient safety events from a large number of patient complaint corpus,so as to find out the factors that affect the safety of patients as soon as possible.The purpose of this study is(1)to form a classification standard of patient safety events suitable for Chinese patient complaints;(2)to form a corpus of patient safety events text;(3)to train a classification model to identify complaints related to safety events.[Methods] The collected 7009 patient complaint texts were manually annotated,and a variety of resampling methods were used to balance the unbalanced patient complaints.The machine learning methods of Random Forest,Support Vector Machine,AdaBoost and Gradient Boosting Decision Tree are used to identify the complaints related to safety incidents.[Results] The results showed that there were 660 complaint texts related to patient safety out of 7009 texts.The experiment on the corpus showed that the greater the imbalance ratio,the worse the recognition results.When SMOTE and Tomek Links Removal balance data was used to 1: 1,the optimal results were obtained by using random forest,including accuracy 99.58%,recall 97.96%,F-Measusre 98.76%,G-mean 98.77%,area under ROC curve 99.86% and area under PR curve 99.90%.[Conclusions] In this paper,a classification standard of patient safety events suitable for Chinese text is proposed.According to this standard,a corpus of patient safety events text is constructed,and a model for identifying complaints related to safety events from patients’ complaints is constructed.The experimental results of the model show that it can meet the actual needs,facilitate the timely intervention of medical institution managers,and helpful to identify the adverse factors before the occurrence of medical disputes or violent incidents and avoid the occurrence of medical disputes.