Intelligent identification of the big data of liver injury-related adverse drug reactions based on text database
10.3969/j.issn.1001-5256.2022.02.024
- VernacularTitle:基于文本知识库的肝损伤药物不良反应大数据智能识别研究
- Author:
Feilin GE
1
,
2
;
Yuming GUO
1
;
Ming NIU
1
;
Xu ZHAO
1
;
Zhaofang BAI
1
;
Jiabo WANG
1
;
Xiaohe XIAO
1
Author Information
1. Department of Hepatology/China Military Institute of Chinese Medicine, The Fifth Medical Center of Chinese PLA General Hospital, Beijing 100039, China
2. School of Chinese Materia Medica, Beijing University of Chinese Medicine, Beijing 100029, China
- Publication Type:Original Articles_Other Liver Diseases
- Keywords:
Liver Diseases;
Drug-Related Side Effects and Adverse Reactions;
Knowledge Bases;
Artificial Intelligence
- From:
Journal of Clinical Hepatology
2022;38(2):387-391
- CountryChina
- Language:Chinese
-
Abstract:
Objective To establish the intelligent identification method for the big data of liver injury-related adverse drug reaction (ADR) based on the construction of text database. Methods With the keywords including "drug-induced liver injury" and "abnormal liver function" and a search time of January 1, 2012 to December 31, 2016, 5% (4152 cases) of the case reports of liver injury-related ADR were retrieved and extracted from the China Adverse Drug Reaction Monitoring System, and then based on clinical reevaluation by physicians, these cases were classified into "negative cases", "suspected cases", and "confirmed cases". On this basis, key elements (including ADR name, biochemical parameter, and clinical symptoms) were identified. An intelligent identification method for liver injury-related ADR was established based on the correlation analysis between key elements and clinical reevaluation and the receiver operating characteristic (ROC) curve for determining cut-off values, and the method of cross validation was used to evaluate the performance of this intelligent identification method. Results The formula for the evaluation and identification of liver injury-related ADR was as follows: total score (M)=symptom score+index score+ADR name score. This formula showed the best discriminatory ability to distinguish "negative case" from "suspected case" or "confirmed case" at M=5 (area under the ROC curve [AUC]=0.97), with a sensitivity of 99.57% and a specificity of 84.61%, and it showed the best discriminatory ability to distinguish "confirmed case" from "suspected case" or "negative case" at M=12 (AUC=0.938), with a sensitivity of 87.93% and a specificity of 85.98%. Conclusion This method provides reference and basis for intelligent identification and evaluation of big data on liver injury-related ADR and is expected to effectively reduce the burden of manual processing of ADR big data and provide effective tools and methodological demonstration for early risk signal identification and warning of liver injury-related ADR.
