Potential confusion risk screening and display-name optimization for Chinese drug names based on a multimodal similarity model
- VernacularTitle:基于多模态相似度模型的中文药品名称潜在混淆风险筛查与显示名优化
- Author:
Feng WANG
1
;
Qi WANG
1
;
Yuanyuan FAN
2
;
Panke ZENG
1
;
Haiyan DONG
1
Author Information
1. Dept. of Pharmacy，The First Affiliated Hospital of Xi’an Jiaotong University，Xi’an 710061，China
2. Dept. of Pharmacy，The First Affiliated Hospital of Xi’an Jiaotong University，Xi’an 710061，China;Dept. of Pharmacy，Northwest Women’s and Children’s Hospital，Xi’an 710061，China
- Publication Type:Journal Article
- Keywords:
drug names;
multimodal similarity model;
naming optimization;
look-alike medicine;
sound-alike medicine;
confusion risk;
linguistic similarity;
visual similarity
- From:
China Pharmacy
2026;37(19):2570-2576
- CountryChina
- Language:Chinese
-
Abstract:
OBJECTIVE To construct a multimodal similarity model for Chinese drug names and design optimized display names，compare the potential confusion risk of display names of the same drug catalogue in different scenarios，so as to provide a quantitative basis for the management of drug display names.METHODS A total of 1 221 drug entries were extracted from the drug dictionary of a tertiary grade A hospital，forming 744 810 drug pairs. “Five rules” were summarized based on the confusion characteristics of candidate confusable drug pairs under the currently used in-hospital display names （Scenario 1） and standard display names （Scenario 2），to generate optimized display names （Scenario 3）. A multimodal similarity model integrating Pinyin，glyph，character sequence and rendered visual features was constructed. The same threshold （total similarity S total ≥0.70） was adopted to screen candidate confusable drug pairs. Confusion risk grading and transition，confusion characteristic labeling，confusion network clustering and residual cluster analysis were carried out. The naming mechanism of optimized display names was investigated through display morphology indicators and rule ablation，and the robustness of the comparison results among the three scenarios was evaluated through similarity threshold and fusion weight sensitivity analysis.RESULTS The numbers of candidate confusable drug pairs in the three scenarios were 156，269 and 83 respectively，among which the number of pairs at high risk and above was 109， 215 and 64 respectively. Compared with Scenario 1，Scenario 3 reduced the number of candidate confusable drug pairs by 46.8%，and the number of those at high risk or above decreased by 41.3%. Display morphology analysis showed that Scenario 3 had the lowest average ink density （0.057 3） and the largest average number of structural slots （2.99）. Rule ablation analysis showed that after disabling the rule of manufacturer characteristic abbreviation prompts，the number of candidate confusable drug pairs increased the most （+74 pairs），while disabling structured naming alone resulted in a net decrease of 9 pairs. The residual clusters mainly involved drugs with the same active ingredient but different specifications or subtypes，as well as series products from the same manufacturer with similar main name stems. Under the threshold setting of 0.60 to 0.80 with a step size of 0.025 and 11 different weight configurations （including the baseline value），the number of candidate confusable drug pairs always maintained the ranking of Scenario 3＜Scenario 1＜Scenario 2.CONCLUSIONS The multimodal similarity model can be used for quantitative screening of potential confusion of Chinese drug display names，and the ranking of the number of candidate confusable drug pairs among scenarios remains stable within the set parameter range. The optimized display names based on the “five rules” can reduce the number of candidate confusable drug pairs， and the number of those at high risk or above in the hospital’s drug catalogue，but their impact on actual identification behavior and clinical medication safety still needs further verification.
