中国血液净化

中国血液净化 ›› 2026, Vol. 25 ›› Issue (06): 501-507.doi: 10.3969/j.issn.1671-4091.2026.06.012

• 临床研究 • 上一篇    下一篇

基于人工智能建立和评估抗肿瘤药物相关急性肾损伤的临床预测模型

李泽薇    张 怡   濮丽天   李 力   杨 萌  熊丽焱   崔健臣   沈 颖   王新宇   邓钦元  徐 剑   

  1. 650032 昆明，云南省第一人民医院1肾内科 2内分泌与代谢科 3麻醉科
    650032 昆明，4昆明理工大学附属医院肾内科
  • 收稿日期:2025-06-16 修回日期:2026-01-19 出版日期:2026-06-12 发布日期:2026-06-12
  • 通讯作者: 徐剑 E-mail:xujian3979@sina.com
  • 基金资助:
    国家自然科学基金(82460150)；云南省万人名医计划(YNWR-MY-2018-019)

Establishment and evaluation of a clinical prediction model for acute renal injury related to anti-tumor drugs using artificial intelligence

LI Ze-wei， ZHANG Y， PU Li-tian， LI Li，YANGXIONG Li-yan， CUI Jian-chen， SHEN Ying， WANG Xin-yu， DENG Qin-yuan，XU Jian   

  1. Department of Nephrology, 2Department of Endocrine and Metabolic Diseases, and 3Department of Anesthesiology, The First People’s Hospital of Yunnan Province, Kunming 650032, China；4Department of Nephrology, The Affiliated Hospital of Kunming University of Science and Technology, Kunming 650000, China
  • Received:2025-06-16 Revised:2026-01-19 Online:2026-06-12 Published:2026-06-12
  • Contact: 650032 昆明，云南省第一人民医院1肾内科；650032 昆明，4昆明理工大学附属医院肾内科 E-mail:xujian3979@sina.com

摘要： 目的 探讨抗肿瘤药物相关急性肾损伤(acute kidney injury，AKI)的危险因素，并建立人工智能(artificial intelligence，AI)预测模型。 方法 选取2015年1月1日—2023年8月1日在云南省第一人民医院确诊的恶性肿瘤患者，根据改善全球肾脏病预后(Kidney Disease:Improving Global Outcomes，KDIGO)临床实践指南标准分为AKI组和非AKI组。采用单因素及多因素Logistic回归分析筛选抗肿瘤药物相关AKI的独立危险因素，并构建决策树、随机森林等机器学习模型。采用受试者工作特征曲线下面积(area under curve,AUC)等指标评估模型性能。 结果 共纳入272例患者，其中AKI组89例，非AKI组183例。单因素Logistic回归分析显示：体质量指数(body mass index,BMI)正常(OR=0.533,95%CI：0.295～0.964，P=0.037)、曾接受放疗(OR=0.369,95%CI：0.147～0.922，P=0.033)或手术的患者(OR=0.142,95%CI：0.065～0.311，P＜0.001)AKI发生风险较低；呼吸系统肿瘤(OR=2.162，95%CI：1.172～3.991，P=0.014)、顺铂治疗(OR=2.135，95%CI：1.178～3.869，P=0.012)和非铂类化疗药物(OR=9.247，95%CI：4.271～20.017，P＜0.001)为独立危险因素。决策树模型AUC=0.660，LASSO回归模型AUC=0.864，神经网络模型AUC=0.833，随机森林模型在所有模型中表现最优，AUC为0.870，召回率为0.919。 结论 基于随机森林算法建立的AI模型可早期识别抗肿瘤药物相关AKI的高危患者，为临床干预提供决策支持。

关键词: 人工智能, 急性肾损伤, 机器学习, 预测模型

Abstract: Objective  This study aimed to investigate the risk factors for acute kidney injury (AKI) associated with antineoplastic drugs and to develop an artificial intelligence (AI)-based prediction model.  Methods A cohort of 272 cancer patients diagnosed and treated at the First People's Hospital of Yunnan Province between January 1, 2015, and August 1, 2023 were enrolled. These patients were divided into the AKI group (n=89) and the non-AKI group (n=183) based on the Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Both univariate and multivariate logistic regression analyses were employed to identify independent risk factors, and machine learning models, such as decision trees and random forests algorithms, were constructed. The performance of these models was assessed using metrics including the area under the receiver operating characteristic curve (AUC). Results Of the 272 patients included, 89 (32.7%) developed AKI. Univariate analysis showed that patients with normal body mass index (BMI) (OR=0.533, 95%CI：0.295～0.964，P=0.037), previous radiotherapy (OR=0.369, 95%CI：0.147～0.922，P=0.033), or surgery (OR=0.142, 95%CI：0.065～0.311，P＜0.001) were associated with a lower risk of AKI. However, multivariate analysis identified respiratory system cancer (OR=2.162, 95%CI：1.172～3.991, P=0.014), the use of cisplatin (OR=2.135, 95% CI: 1.178～3.869, P=0.012), and non-platinum chemotherapy drugs (OR=9.247, 95% CI：4.271～20.017, P＜0.001) as independent risk factors. The decision tree model achieved an AUC of 0.660, the LASSO regression model an AUC of 0.864, and the neural network model an AUC of 0.833. Among all models, the random forest model exhibited the optimal predictive performance, with an AUC of 0.870 and a recall of 0.919.  Conclusions The AI model utilizing the random forest algorithm enables early identification of high-risk patients for AKI associated with antineoplastic drugs, thereby providing valuable decision support for clinical intervention.

Key words: Artificial intelligence, Acute kidney injury, Machine learning, Predictive modeling

中图分类号: