关键词: font-size:medium, ">数据挖掘；辅助用药；自然语言处理；Meta分析

Abstract: Objective： Using natural language processing (NLP) technology to analysis off-label drug use. Methods： The study used self-established drug instruction database to train deep learning model (Bi-directional Long Short-Term Memory Network-CRF， BiLSTM-CRF) to identify disease entities， combined with the data provided by National Population and Health Science Data Sharing Service Platform to find the off-label drug use， and applied Meta-analysis to verify the rationality of the corresponding adjuvant medication. Results： The F-score of the BiLSTM-CRF model was 84.19%. The results showed that 47 kinds of suspected off-label drug use were obtained， and 10 kinds of off-label drug use were found by manual screening. Meta-analysis verified the rationality of Shenmai injection in adjurant treatment of acute myocardial infarction. Conclusion： This study effectively extracted the disease entity in the drug label， and proved that NLP played a major role in the transformation of the medical data from unstructured data into structured data， which is of great significance for discovering the off-label drug use and promoting rational drug use.

Key words: font-size:medium, ">Data mining； Adjuvant drugs； Nature language processing； Meta-analysis