文献速递细菌感染、耐药与抗菌药物

确定抗生素处方适应症以促进抗菌药物管理的方法:范围综述

Methods for determining indications for antibiotic prescriptions to facilitate antimicrobial stewardship: a scoping review.
PubMed 文献 · Clin Microbiol Infect · 2026-05-25

多数研究来自高收入国家、多中心、回顾性设计;57%使用自动化方法(多为基于本体的标准化编码系统),43%手动提取,仅3%使用规则或人工智能驱动方法。适应症注册方法异质且各有局限,需标准化定义和可扩展AI模型。

要点

  • 多数研究采用自动化方法确定适应症,但规则和AI驱动方法仅占3%。
  • 手动提取详细但资源密集;结构化编码缺乏粒度。
  • 需标准化定义和可扩展AI模型以精确监测抗生素使用。

核心结论

  • 确定抗生素处方适应症的方法异质且存在明显局限,需标准化定义和可扩展AI模型以实现精确可持续的抗生素使用监测。

主要发现

  • 多数研究在高收入国家开展,涉及多中心,采用回顾性设计。
  • 43%的研究报告手动提取适应症,57%使用自动化方法,最常用的是基于本体的标准化医学编码系统,其次是处方者注册的适应症。
  • 规则驱动和人工智能驱动方法仅用于3%的研究。
  • 主要数据源类型为结构化数据集,其次是电子健康记录数据、临床笔记和账单或保险记录。
  • 大多数研究报告多个感染类别(如呼吸道感染),超过一半提供感染级粒度(如咽炎或肺炎)。

关键数值与推荐

  • 手动方法详细但资源密集。
  • 结构化编码缺乏粒度。
  • 处方者注册系统捕获治疗意图适应症,但面临工作流程和可用性障碍,限制准确性和采用。
  • 规则驱动和AI驱动方法仅很少用于抗菌药物管理目的识别适应症。

临床怎么用

  • 临床可参考自动化方法(如基于本体的标准化编码系统)确定抗生素处方适应症,但需注意结构化编码缺乏粒度。
  • 手动提取适应症虽详细但资源密集,需权衡资源与准确性。
  • 处方者注册系统需克服工作流程和可用性障碍以提高准确性和采用率。

局限性

  • 适应症注册方法异质且存在明显局限。
  • 手动方法资源密集。
  • 结构化编码缺乏粒度。
  • 处方者注册系统面临工作流程和可用性障碍,限制准确性和采用。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: Effective antimicrobial stewardship (AMS) requires accurate information on the reason for prescribing antimicrobials. To design targeted interventions aimed at improving guideline adherence, accurate information on the indication for each prescription is necessary. This scoping review maps the existing methods used for systematically determining indications for antibiotic prescriptions. SOURCES: A search in Medline and Embase was performed until 1 October 2025. Studies determining indications of antibiotic prescriptions for the purpose of AMS were included. CONTENT: A total of 157 studies were identified. Most were conducted in high-income countries, involved multiple centres, and used retrospective designs. Manual indication extraction was reported in 43% of studies, whereas 57% used automated methods, most commonly ontology-based systems using standardized medical coding schemes, followed by prescriber-registered indications. Rule-based and artificial Intelligence (AI)-driven methods were only used in 3% of studies. The predominant data source types were structured datasets, followed by electronic health record data, clinical notes, and billing or insurer records. Most studies reported multiple infection categories (e.g. respiratory tract infection), and over half provided infection-level granularity (e.g. pharyngitis or pneumonia). IMPLICATIONS: Indication registration approaches were heterogeneous and with notable limitations. Manual approaches are detailed but resource intensive. Regarding automated methods, structured codes lack granularity, and prescriber-registered systems capture the intended indication for therapy but face workflow and usability barriers that limit accuracy and adoption. Rule-based and AI-driven methods are only rarely used to identify indications for AMS purposes. Standardized definitions, improved data capture, and scalable AI models that leverage both structured data and unstructured clinical text are required to enable precise and sustainable monitoring of antibiotic use.

来源:Clin Microbiol Infect,查看原文。版权归原作者所有。

引用

  1. Kuijpers SME, Siepel M, Schut MC, et al. Methods for determining indications for antibiotic prescriptions to facilitate antimicrobial stewardship: a scoping review. Clin Microbiol Infect. 2026. PMID: 42190950. doi:10.1016/j.cmi.2026.05.021

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。