目的 本研究旨在系统评估人工智能（artificial intelligence,AI）工具在医保双通道外配处方审核中的应用效能,探索人机协同审核模式的实践价值。方法 回顾性选取2025年1月—3月本院医院信息系统（hospital information system,HIS）中138份医保双通道外配处方作为研究对象,运用腾讯元宝、豆包、Kimi Chat及DeepSeek网页版四种AI工具,依据《审核规范》和《医院处方点评管理规范（试行）》对处方规范性及用药适宜性进行筛查,以人工复核结果为金标准。比较4种AI工具对处方点评的时效性、准确率和问题检出一致性。结果 腾讯元宝（90±12 s）、豆包为（36±4 s）、Kimi Chat为（20±5 s）的平均审核耗时均显著低于人工复核（276±23 s）,其中Kimi Chat的审核速度最快。腾讯元宝、豆包和Kimi Chat假阳性率分别为9.23%、3.08%和3.08%,假阴性率分别为62.5%、12.50%和75%,豆包准确率最高。豆包检出率最高（62.50%）,腾讯元宝检出率次之（25.00%）,Kimi Chat检出率最低（12.50%）。结论 AI工具在医保双通道外配处方审核中时效性优于人工,豆包一致性检出率和准确率最高,具有较好的应用效能,人机协同模式值得关注。

Objective The purpose of this study is to systematically evaluate the efficacy of artificial intelligence (AI) tools in the review of dual-channel outpatient prescriptions for health insurance, and to explore the practical value of the human-machine collaborative review model. Method Retrospectively selected 138 Medicare dual-channel outpatient prescriptions in our hospital information system (HIS) from January to March 2025 as the study object and utilized four AI tools, namely, Tencent Yuanbao (deep seek), Doubao, Kimi Chat, and DeepSeek web version, to screen the prescriptions for standardization and medication appropriateness based on the Audit Specification and the Hospital Prescription Review Management Code (for Trial Implementation) to screen prescriptions for standardization and medication appropriateness, with manual review results as the gold standard. Comparison of timeliness, accuracy, and problem detection consistency of prescription reviews by 4 AI tools. Result The average review time consumed by Tencent Yuanbao (90±12 s), Doubao for (36±4 s), and Kimi Chat for (20±5 s) was significantly lower than that of manual review (276±23 s), with Kimi Chat being the fastest (P<0.05). Tencent Yuanbao (deep seek), Doubao and Kimi Chat had false-positive rates of 9.23%, 3.08% and 3.08%, respectively, and false-negative rates of 62.5%, 12.50% and 75%, respectively, with Doubao having the highest accuracy rate. Doubao (Deep Thinking) achieved the highest detection rate (62.50%), followed by Tencent Yuanbao (deep seek) with a detection rate of 25.00%, while Kimi Chat had the lowest detection rate at 12.50% (P<0.05). Conclusion AI tools in the medical insurance dual-channel external dispensing prescription audit timeliness is better than manual, beanbag consistency detection rate and accuracy rate is the highest, has a better application efficacy, human-computer synergy mode is worth paying attention to.