目的: 探讨在药品价格监测工作中智能化提取和测算中成药日均用量的方法及应用,为提升药品价格监测的精细化水平和科学性提供关键数据支持。方法: 以深圳药品交易平台运用人工智能（AI）技术提取中成药日均用量为实践案例,针对中成药剂型复杂、说明书文本多场景描述及日均用量传统人工提取困难等难题,平台基于DeepSeek开源框架大语言模型,构建了融合自然语言处理（NLP）、规则引擎与双重异常检测算法的智能系统。通过NLP精准提取药品说明书中的剂量、频次和单位信息,结合规则引擎实现多场景分割与精准日均用量计算,并利用双重异常检测算法反向校验并修正数据。结果: 实践表明,该智能系统对1.2万条中成药用法用量数据的日均用量提取正确率达95.51%,双重异常检测算法的F1值达0.9413,说明系统具有良好的中成药用法用量语言处理和自我异常检测等综合能力,提取效率较传统人工提升12倍,且通过轻量化本地部署,保障了数据安全。结论: 研究验证了AI技术在计算中成药差比价、识别隐性价格操纵、支撑药品价格动态监测及编制中成药价格指数上可实操的技术路径,为整个行业的智能化、规范化发展做了初步尝试和探索。

Objective: This research explores the method and application of intelligent extraction and calculation of daily average dosage for Chinese patent medicines in drug price monitoring, to provide key data support for improving the refinement and scientificity of drug price monitoring. Methods: In view of the difficulties of complex dosage forms of Chinese patent medicines, multi-scenario descriptions in package inserts, and the difficulty of traditional manual extraction of daily dosage, an intelligent system was developed on the DeepSeek open-source framework, taking the application of artificial intelligence (AI) technology to extract the daily average dosage of traditional Chinese patent medicines on the Shenzhen Drug Trading Platform as a practical case. This system synergistically employs natural language processing (NLP), rule engine, and dual anomaly detection algorithm. NLP is utilized for the precise extraction of dosage, frequency, and unit information from the package inserts. The rule engine enables multi-scenario segmentation and accurate daily dosage calculation. The dual anomaly detection algorithm provides a reverse validation mechanism to correct data inconsistencies. Results: Practice shows that the intelligent system attained an accuracy rate of 95.51% when processing 12000 Chinese patent medicine usage and dosage records. The F1 value of the dual anomaly detection algorithm reaches 0.9413, indicating that the system has good comprehensive capabilities such as language processing of traditional Chinese patent medicine usage and dosage, and self anomaly detection. Extraction efficiency is 12 times higher than traditional manual processing. The lightweight local deployment solution concurrently ensured data security. Conclusion: This study validates that AI technology provides practical technical solutions for calculating price differences of Chinese patent medicines, identifying covert price manipulation, supporting dynamic drug price monitoring, and compiling price indices of Chinese patent medicines. It represents initial exploratory efforts toward intelligent and standardized development in the pharmaceutical industry.