基于BERT的医疗安全事件智能分类研究与实践

Purpose/Significance To improve the manual mode of medical safety event classification and evaluation, and improve work efficiency and timeliness. Method/Process Select past medical safety event data for preprocessing, use BERT (Bidirectional Encoder Representations from Transformers)model for training, testing, and iterative optimization, and construct an intelligent classification and prediction model for medical safety events. Result/Conclusion The model was used to classify 466 medical safety incidents reported by clinical departments from January to November 2022, including 267 medical adverse events (including 25 Level I events, 105 Level II events, 36 Level III events, and 101 Level IV events) and 199 patient related safety hazards, F1 value reaches 0.66. Applying BERT to assist in the classification and evaluation of medical safety incidents can improve the efficiency and timeliness of this work to a certain extent, and help to intervene in medical safety risk hazards in a timely manner.