Extracting Safety-II Factors From an Incident Reporting System by Text Analysis

Introduction The use of electric health records (EHRs) has spread worldwide and has helped record huge amounts of data. However, despite accumulated data from EHRs, especially text data, the information has been underutilized. Our research questions and aims are as follows: How can an incident report system extract common themes behind incidents, good practices, improved quality, and safety based on the Safety-II/resilient healthcare approach? Methods We extracted data from the electronic incident reporting system of the Yokohama City University Medical Center between April 1, 2016 and March 31, 2018. We utilized natural language processing and text mining to extract concept categories and word patterns. We also used the incident levels as outcomes, as well as classification and regression tree analysis to obtain associated text combinations. Results A total of 17,231 cases were reported through the electronic incident reporting system in our hospital during the study period. Hospital staff has to be prepared for incidents with complex mechanisms in daily practice. The hospital staff tend to focus on individual actions rather than considering a systematic approach. Conclusion Certain combinations of professions and contents may contribute to resilient management. Studies on Safety-II management utilizing clinical information and text records are needed.