Investigation and analysis on data mining problems in adverse drug reaction signal detection study in China
10.3760/cma.j.cn114015-20230905-00661
- VernacularTitle:国内药物不良反应信号检测文献中数据挖掘问题的调查和分析
- Author:
Rui DAI
1
;
Qingxia ZHANG
;
Yang HU
;
Hao XIE
;
Huanling WANG
;
Bin ZHAO
Author Information
1. 中国医学科学院北京协和医学院北京协和医院药剂科，北京 100730
- Publication Type:Journal Article
- Keywords:
Pharmacovigilance;
Drug-related side effects and adverse reaction;
Adverse drug reaction reporting systems;
Signal detection;
Data mining;
Evaluation studi
- From:
Adverse Drug Reactions Journal
2023;25(12):717-723
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To understand the problems of data mining in adverse drug reaction (ADR)/adverse event (AE) signal detection study in China.Methods:The literature on ADR/AE signal detection study in SinoMed, CNKI, WanFang Data and VIP databases were retrieved (up to May 30, 2022). The relevant content of the data mining in the literature was investiagted and evaluated from the following 4 dimensions and 9 items: (1) background data, including 1 item; (2) data preprocessing, including 4 items such as drug mapping, AE mapping, missing value processing, and data deduplication; (3) data mining algorithm (DMA), including 3 items such as DMA selection, DMA formula interpretation, and signal threshold; (4) interpretation of the results, including 1 item. According to the relevant specifications and technical requirements of data mining, the reporting/reporting error-free rate of the 4 dimensions and 9 items in the literature was taken as the overall quality evaluation index. Reporting/reporting error-free rate ≥60% was considered to be an excellent level of overall quality.Results:A total of 165 articles were included. On the background data dimension, the reporting/reporting error-free rate of using all the other drug data of the entire database in the literature was 35.2% (58/165), which did not reach an excellent level. On the data preprocessing dimension, the reporting/reporting error-free rates of drug mapping, AE mapping, missing value processing, and data deduplication in the literature were 22.4% (37/165), 73.9% (122/165), 10.3% (17/165), and 55.2% (91/165), respectively. The reporting/reporting error-free rate on this dimension was 40.5% (267/660), which did not reach the excellent level, only the rate of AE mapping reached the excellent level. On the DMA dimension, the reporting/reporting error-free rates of ≥2 DMA, DMA formula interpretation, and signal threshold in the literature were 63.6% (105/165), 78.2% (129/165), and 87.9% (145/165), respectively. The reporting/reporting error-free rate on this dimension was 76.6% (379/495), which reached an excellent level. The reporting/reporting error-free rate was 87.4% (144/165), reaching an excellent level. The signals were interpreted as "positive" or "negative" signals in 7 articles, and the meaning of signals were interpreted as causality in 14 articles. The overall reporting/reporting error-free rate in the 165 literature, analyzed from 9 items on 4 dimensions, was 57.1% (848/1 485), which did not reach the excellent level.Conclusion:The main problems in the domestic literature of ADR/AE signal detection study are the incomplete selection of background data and the lack of data preprocessing, suggesting that the further relevant studies in China should be improved on above 2 dimensions for better quality of ADR/AE signal detection research.
