文献速递细菌感染、耐药与抗菌药物医院感染防控

人工智能在抗菌药物管理中的应用:预测、临床实践与实施挑战

Artificial intelligence in antimicrobial stewardship: prediction, clinical applications, and implementation challenges.
PubMed 文献 · Clin Microbiol Infect · 2026-07-28

本文综述人工智能在抗菌药物管理中的应用,涵盖耐药预测、经验性治疗选择、诊断管理和抗菌药物优化。AI模型可支持风险分层和个体化决策,但存在预测目标异质、数据泄漏和外部验证有限等方法学局限。成功实施依赖工作流程整合、可解释性、治理和临床医生采纳,AI应作为增强管理专业能力的工具而非替代临床判断。

要点

  • AI模型可基于电子健康记录预测耐药和多重耐药感染风险,支持经验性治疗选择。
  • AI应用于诊断管理,包括优化血培养使用、分子诊断合理化和结果解读支持。
  • 实施挑战包括工作流程整合、可解释性、治理和临床医生采纳,需前瞻性多中心评估。

核心结论

  • AI有潜力通过更精准、可扩展和工作流程整合的决策支持来增强抗菌药物管理,但应视为增强管理专业能力的工具而非替代临床判断。

研究设计与样本

  • 综述评估了AI和机器学习方法用于耐药预测、诊断管理、抗菌药物优化和临床实施的相关研究。

主要发现

  • AI模型已开发用于预测耐药和识别多重耐药感染风险患者,可能支持经验性治疗选择和患者层面风险分层。
  • AI应用延伸至诊断管理,包括优化血培养使用、分子诊断合理化和微生物结果解读支持。
  • 在治疗阶段,AI可能支持降阶梯、静脉转口服、疗程再评估和管理审查优先级排序。

临床怎么用

  • AI可整合临床、微生物和情境数据,支持更及时和个体化的决策以优化抗菌药物使用。
  • 成功实施取决于模型性能、临床工作流程整合、可解释性、治理和临床医生采纳。
  • 未来工作应优先前瞻性和多中心评估,以及解决安全性、透明度、公平性和临床监督的治理框架。

局限性

  • 重要方法学局限包括预测目标异质、数据泄漏和有限的外部验证。

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: Antimicrobial resistance (AMR) continues to threaten modern infectious diseases practice. Antimicrobial stewardship programmes (ASPs) remain central to optimizing antimicrobial use, yet stewardship has become increasingly challenging because of rising clinical complexity, expanding data sources, and persistent workforce and analytic constraints. Artificial intelligence (AI) may strengthen stewardship by integrating clinical, microbiologic, and contextual data to support more timely and individualized decision-making. OBJECTIVES: To review current applications of AI in antimicrobial stewardship, with emphasis on resistance prediction and risk stratification, empiric therapy selection, diagnostic stewardship, antimicrobial optimization, and implementation challenges in clinical practice. SOURCES: Relevant studies evaluating AI and machine-learning approaches for AMR prediction, diagnostic stewardship, antimicrobial optimization, and clinical implementation were reviewed. CONTENT: AI-based models have been developed to predict AMR and identify patients at risk for multidrug-resistant infections using electronic health record data. These approaches may support empiric therapy selection and patient-level risk stratification, although important methodological limitations remain, including heterogeneous prediction targets, data leakage, and limited external validation. AI applications also extend to diagnostic stewardship, including optimization of blood culture use, rationalization of molecular diagnostics, and support for interpretation of microbiologic results. In the therapeutic phase, AI may support de-escalation, intravenous-to-oral conversion, duration-of-therapy reassessment, and prioritization of stewardship review, aligning clinical decisions with antimicrobial stewardship goals. However, successful implementation depends not only on model performance but also on effective integration into clinical workflows, interpretability, governance, and clinician uptake. IMPLICATIONS: AI has the potential to enhance antimicrobial stewardship by enabling more precise, scalable, and workflow-integrated decision support. Future work should prioritize prospective and multicentre evaluation, careful implementation in routine care, and governance frameworks that address safety, transparency, equity, and clinician oversight. AI should be viewed as a tool to augment ASP expertise rather than replace clinical judgement.

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

引用

  1. Matsuo T, Nigo M, Borgonovo F, et al. Artificial intelligence in antimicrobial stewardship: prediction, clinical applications, and implementation challenges. Clin Microbiol Infect. 2026. PMID: 42521015. doi:10.1016/j.cmi.2026.07.032

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。