Abstract: Background　Intelligent assistance (AI) systems are gradually being introduced into primary healthcare practice to address issues such as the lack of experience among doctors in primary healthcare institutions and low diagnostic accuracy，aiming to improve the efficiency and accuracy of medical services. However， systematic research is scant on the specific impact of AI systems on the clinical diagnosis by grassroots doctors. Objective　To explore the impact of AI prompts on the clinical diagnosis by grassroots doctors，providing decision-making support in AI-assisted improvement of their clinical diagnosis. Methods　Data were extracted from the electronic prescription record platform of primary medical institutions in Anhui Province，including all diagnostic modification records made by doctors in all township-level medical health institutions （service centers/health centers） and village-level medical health institutions （service stations/clinics） across Anhui province from July 2020 to November 2021，prompted by the intelligent assistance system. Records were exported after desensitization. Major analyses included the number of modified diagnoses，modification rate，common final diagnoses，and the composition of initial diagnoses for typical final diagnoses （coronary heart disease，uterine hemorrhage，tuberculosis，and bacterial dysentery）. Village health clinics or community health service stations were defined as primary institutions，while township hospitals or community health service centers were defined as secondary institutions. The number of modified diagnoses, modification rate, person-times and ordering of common final diagnoses, and initial diagnoses for typical final diagnoses and the composition were compared in different geographical locations，different months，different systemic diseases，and different types of health institutions. Results　This study extracted a total of 1.3636 million modified diagnoses，with an overall incidence of modified diagnoses of 2.02%（1.363 6 million / 67.413 8 million），showing a trend of fluctuation or slight decline. The modification rate in northern，southern，and central parts of Anhui province was 2.33%（641 900/27 602 500），2.01%（288 900/14 380 700），and 1.70%（432 800/25 430 500），respectively. The modification rate in secondary and primary institutions was 2.25%（901 600/40 023 400）and 1.69%（462 000/27 390 400），respectively. The highest modification rate of diagnoses for common systemic diseases was seen in the endocrine system（2.39%），followed by the circulatory system（2.20%），and lowest in the skin and immune system（0.22%）. The top 20 most frequently modified diagnoses accounted for 66.11% of all modifications （901 500 / 1 363，600）. The initial diagnoses of typical final diagnoses were generally dominated by certain specific systemic diseases. Conclusion　AI-assisted diagnosis has a significant impact on the daily clinical services of primary care physicians and deserves high attention and systematic evaluation.

Key words: Artificial intelligence, Grassroots doctors, Primary healthcare institutions, Intelligent assistance, Diagnosis modification

摘要： 背景　为应对基层医疗机构医生经验不足和诊断准确率低等问题，智能辅助系统逐渐被引入基层医疗实践，旨在提升医疗服务的效率和准确性。然而，目前关于智能辅助系统对基层医生临床诊断的具体影响仍缺乏系统性的研究。目的　探讨智能提示对基层医生临床诊断的影响，为利用人工智能技术改善基层医生临床诊断提供决策依据。方法　本研究数据提取自安徽省基层医疗机构的电子处方记录平台，包括2020年7月—2021年11月全省全部乡级医疗卫生机构（服务中心/卫生院）和村级医疗卫生机构（服务站/卫生室）的医生在智能辅助系统提示下所做的全部诊断修改记录，记录经过脱敏后导出。主要分析：修改诊断次数、修改率、常见最终诊断、典型最终诊断（冠心病、子宫出血、结核病和细菌性痢疾）的初始诊断的构成。将村卫生室或社区卫生服务站简称一级机构，乡镇卫生院或社区卫生服务中心简称二级机构。比较不同地理位置、不同月份、不同系统疾病、不同类型卫生机构的修改诊断次数及修改率、常见最终诊断的人次数与排序、典型最终诊断的初始诊断及其构成。结果　本研究共提取136.36万人次的修改诊断，接诊人次数为6 741.38万人次，总体修改率为2.02%（136.36万/6 741.38万），并呈“波动或微弱下降”的趋势。省北、省南、省中的修改率分别为2.33%（64.19万/2 760.25万）、2.01%（28.89万/1 438.07万）、1.70%（43.28万/2 543.05万）；二级机构和一级机构的修改率分别为2.25%（90.16万/4 002.34万）和1.69%（46.20万/2 739.04万）。常见系统疾病的诊断修改率以内分泌系统为最高（2.39%），其次为循环系统（2.20%），皮肤免疫系统最低（0.22%）。前20位修改诊断次数占全部修改次数的66.11%（90.15万/136.36万）。典型最终诊断的初始诊断均以某些特定系统疾病为主导。结论　智能辅助诊断对基层医生的日常诊疗服务产生了重要影响，值得高度重视和系统评估。

关键词: 人工智能, 基层医生, 基层医疗机构, 智能辅助, 修改诊断