中国血液净化

中国血液净化 ›› 2026, Vol. 25 ›› Issue (06): 495-500.doi: 10.3969/j.issn.1671-4091.2026.06.011

• 临床研究 • 上一篇    下一篇

Logistic回归模型与人工神经网络模型对预测维持性血液透析患者频发性透析中低血压的危险因素分析

周 慧   李红芍   蔡玲琍   张飞金   朱 凤   

  1. 325000 温州，1温州市人民医院血液透析中心
  • 收稿日期:2025-08-14 修回日期:2026-01-27 出版日期:2026-06-12 发布日期:2026-06-12
  • 通讯作者: 周慧 E-mail:Zhouhui_d@163.com
  • 基金资助:
    2022年温州市基础性科研项目(Y20220616)

Logistic regression model and artificial neural network model for predicting the risk factors of frequent intradialytic hypotension in maintenance hemodialysis patients

HOU Hui, LI Hong-shao, CAI Ling-li, ZHANG Fei-jin, ZHU Feng   

  1. Hemodialysis Center, Wenzhou People's Hospital, Wenzhou 325000, China
  • Received:2025-08-14 Revised:2026-01-27 Online:2026-06-12 Published:2026-06-12
  • Contact: 325000 温州，1温州市人民医院血液透析中心 E-mail:Zhouhui_d@163.com

摘要： 目的 探究维持性血液透析(maintenance Hemodialysis，MHD)患者发生频发性透析中低血压(intradialytic Hypotension，IDH)的危险因素，比较Logistic回归模型与人工神经网络(artificial neural network，ANN)模型的预测效能。方法 回顾性分析2022年1月─2024年6月232例MHD患者临床资料，按7:3随机分为训练集(n=162)和测试集(n=70)。依据IDH发生率是否≥20%，将训练集患者分为频发IDH组(n=43)和非频发IDH组(n=119)。采用单因素分析比较组间差异，多因素Logistic回归和ANN模型分析危险因素，受试者工作特征(receiver operating characteristic，ROC)曲线下面积(area under curve，AUC)评估模型效能。 结果 频发IDH组年龄(t=4.486，P＜0.001)、透析龄(t=4.813，P＜0.001)、平均超滤量(ultrafiltration volume，UFV)(t=4.802，P＜0.001)及UFV与干体质量比值(UFV to weight ratio，UFV/W)(t=6.218，P＜0.001)均高于非频发组，而高密度脂蛋白(t=4.389，P＜0.001)和血红蛋白(t=4.253，P＜0.001)较低。Logistic回归显示：透析龄(OR=1.031，95%CI:1.002～1.061,P=0.033)、平均UFV(OR=3.865，95%CI:1.498～9.974，P=0.005)、平均UFV/W(OR=2.355，95%CI:1.582～3.506，P＜0.001)是患者频发IDH危险因素，血红蛋白(OR=0.956,95%CI:0.932～0.981,P＜0.001)和高密度脂蛋白(OR=0.120，95%CI:0.022～0.475，P=0.004)是保护因素。Logistic回归模型AUC为0.906，ANN模型AUC为0.907，两者差异无统计学意义(Z=0.238，P=0.812)。测试集验证结果稳健(Logistic AUC=0.892，ANN AUC=0.885，Z=0.152，P=0.879)。 结论 透析龄、超滤负荷(UFV/W＞3.5%)是频发IDH的危险因素，血红蛋白和高密度脂蛋白是保护因素。Logistic与ANN模型预测效能相当，前者解释性强，后者能捕捉复杂非线性关系。

关键词: 维持性血液透析, 频发性透析中低血压, Logistic回归模型, 人工神经网络模型, 预测效能

Abstract: Objective  To investigate the risk factors of frequent intradialytic hypotension (IDH) in patients undergoing maintenance hemodialysis (MHD), and to compare the predictive performance between logistic regression model and artificial neural network (ANN) model.  Methods  A retrospective analysis was conducted on the clinical data of 232 MHD patients treated from January 2022 to June 2024. They were randomly divided into a training set (n=162) and a testing set (n=70) with a ratio of 7:3. Based on the presence of IDH ≥20% as frequent IDH, patients in the training set were categorized into a frequent IDH group (n=43) and a non-frequent IDH group (n=119). Univariate analysis was performed to compare the differences between the two groups. Multifactorial logistic regression model and ANN model were used to analyze the risk factors for frequent IDH. The area under the curve (AUC) of receiver operating characteristic (ROC) was used to assess model performance.  Results  The frequent IDH group had significantly older age (t=4.486, P＜0.001), longer dialysis duration (t=4.813, P＜0.001), higher average ultrafiltration volume (UFV) (t=4.802, P＜0.001) and higher UFV to dry body weight ratio (UFV/W) (t=6.218, P＜0.001) but had significantly lower levels of high-density lipoprotein (HDL) (t=4.389, P＜0.001) and hemoglobin (t=4.253, P＜0.001), as compared with those in the non-frequent IDH group. Logistic regression showed that dialysis duration (OR=1.031, 95%CI:1.002～1.061, P=0.033), average UFV (OR=3.865, 95%CI:1.498～9.974, P=0.005) and average UFV/W (OR=2.355, 95% CI:1.582～3.506, P＜0.001) were the risk factors for frequent IDH, while hemoglobin (OR=0.956, 95% CI:0.932～0.981, P＜0.001) and HDL (OR=0.120, 95% CI:0.022～0.475, P=0.004) were the protective factors for frequent IDH. The AUC was 0.906 for logistic regression model and was 0.907 for ANN model, showing no statistically significant difference (Z=0.238, P=0.812). The results of testing set validation were robust (logistic AUC=0.892, ANN AUC=0.885; Z=0.152, P=0.879).  Conclusion  Dialysis duration and ultrafiltration load (UFV/W＞3.5%) are the risk factors, while hemoglobin and HDL are the protective factors for frequent IDH. The predictive performance of logistic regression model and ANN model is comparable; logistic regression model offers stronger interpretation ability, while ANN model can capture complex nonlinear relationships.

Key words: Maintenance hemodialysis, Frequent intradialytic hypotension, Logistic regression model, Artificial neural network model, Predictive efficacy

中图分类号: