Page 115 - 《中国药房》2026年1期
P. 115
·智慧药学·
机器学习在药物不良反应预测中的应用进展
Δ
1, 2 #
1, 2
许梦佳 1, 2* ,宋 林 ,杨婷婷 ,黄晨蓉 (1.苏州大学附属第一医院药学部,江苏 苏州 215006;2.苏州大学
1
药学院,江苏 苏州 215100)
中图分类号 R969.3;TP181 文献标志码 A 文章编号 1001-0408(2026)01-0105-06
DOI 10.6039/j.issn.1001-0408.2026.01.19
摘 要 药物不良反应(ADRs)是药物出现有害的或与用药目的无关的反应,可导致疾病进程加快、患者住院时间延长等诸多问
题。传统ADRs监测(如自发呈报系统)存在上报率低、数据质量参差不齐等问题,这制约了ADRs的早期防控。随着信息技术的
飞速发展,机器学习凭借其强大的特征挖掘能力和动态时序分析能力,为临床ADRs的管理与决策提供了强有力的支持。本文通
过梳理近年来国内外的相关文献,对机器学习在ADRs预测中的应用进展进行了归纳总结。结果显示,机器学习已逐渐应用于肾
脏、肝脏、心脏及骨髓等靶器官ADRs(如急性肾损伤、药物性肝损伤等)的早期预警和风险预测;虽然机器学习在ADRs预测领域
表现出巨大的应用潜力,但是仍存在临床数据质量控制不足、模型性能评价标准缺失、模型可解释性不足与临床转化困难等局限。
未来,机器学习在ADRs预测领域的发展趋势应遵循“技术-验证-整合”途径,系统性地推动模型落地。
关键词 机器学习;药物不良反应;预测;药物安全;药物警戒
Advances in the application of machine learning in the prediction of adverse drug reactions
1, 2
1, 2
1, 2
1
XU Mengjia ,SONG Lin ,YANG Tingting ,HUANG Chenrong (1. Dept. of Pharmacy, the First Affiliated
Hospital of Soochow University, Jiangsu Suzhou 215006, China;2. College of Pharmaceutical Sciences,
Soochow University, Jiangsu Suzhou 215100, China)
ABSTRACT Adverse drug reactions (ADRs) refer to harmful or unintended reactions unrelated to the intended purpose of
medication administration, which can lead to various issues such as accelerated disease progression and prolonged hospitalization.
Traditional ADRs monitoring systems (such as spontaneous reporting systems) suffer from limitations such as low reporting rates
and inconsistent data quality, which hinder the early prevention and control of ADRs. With the rapid development of information
technology, machine learning has emerged as a powerful tool for management and decision-making of ADRs by leveraging its
strengths in feature extraction and dynamic temporal pattern analysis. By reviewing relevant literature at home and abroad in recent
years, this paper summarizes the progress in the application of machine learning for ADRs prediction. It is found that machine
learning has gradually been applied to the early warning and risk prediction of ADRs in target organs such as the kidneys, liver,
heart and bone marrow (such as acute kidney injury, drug-induced liver injury, and so on). Although machine learning
demonstrates significant application potential in the field of ADRs prediction, it still faces limitations such as inadequate quality
control of clinical data, lack of standardized criteria for model performance evaluation, insufficient model interpretability and
difficulties in clinical translation. In the future, the development trend of machine learning in the field of ADRs prediction should
follow a “technology-validation-integration” pathway to systematically promote the practical implementation of models.
KEYWORDS machine learning; adverse drug reactions; prediction; drug safety; pharmacovigilance
[1]
根据世界卫生组织国际药物监测合作中心规定,药 所出现的有害的或与用药目的无关的反应 。1999-
物不良反应(adverse drug reactions,ADRs)是指以正常 2024年,我国ADRs报告累计2 587.2万份,其中2024年
剂量药物用于预防、诊断、治疗疾病或调节生理机能时 国家 ADRs 监测系统收到新的、严重的 ADRs/不良事件
[2]
报告达90.9万份,较2023年同比增长9.1% 。在此背景
Δ 基金项目 苏州市科技发展计划项目(No.SKY2022135) 下,ADRs所导致的公共卫生问题日益凸显,引发了一系
*第一作者 硕士研究生。研究方向:个体化药物治疗、机器学习。
[3]
列复杂的临床问题,也增加了医疗成本 。
E-mail:xmj19883124@163.com
自发呈报系统是目前我国广泛使用的ADRs监测体
# 通信作者 主任医师,硕士生导师,博士。研究方向:个体化药物
治疗。E-mail:chrishuangcr@163.com 系,该系统能够覆盖广泛的用药人群,并在真实世界中
中国药房 2026年第37卷第1期 China Pharmacy 2026 Vol. 37 No. 1 · 105 ·

