基于机器学习构建血液透析患者透析中高血压风险预测模型-东南大学学报医学版

基于机器学习构建血液透析患者透析中高血压风险预测模型

单位：1. 开滦总医院儿科, 河北唐山 063000;
2. 唐山市工人医院神经内科, 河北唐山 063000

分类号：R459.5;R544.1;R195.1

出版年·卷·期（页码）：2026·45·第三期（365-378）

摘要：

目的: 探讨不同机器学习(ML)算法模型对血液透析患者透析中高血压风险的预测效能,以期为临床早期识别高风险患者及实施个体化干预提供决策支持。方法: 回顾性选取2023年8月至2025年8月期间于开滦总医院进行血液透析的455例终末期肾病患者为研究对象。采用R软件将所有数据集按7∶3比例随机划分为训练集(n=319)与测试集(n=136)。根据透析过程中是否并发高血压,将训练集划分为并发组(n=130)和未并发组(n=189)。基于训练集采用逻辑回归(LR)、决策树(DT)、极端梯度提升(XGBoost)构建血液透析患者透析中高血压风险模型。基于测试集采用曲线下面积(AUC)、灵敏度、特异度及准确率进行评价,并采用决策曲线分析(DCA)评估最优模型的临床净获益。最后,应用沙普利加和解释法(SHAP)对最优模型的预测结果进行可视化解释。结果: 模型性能评估结果显示,基于9个临床变量构建的XGBoost模型在测试集上对血液透析患者透析中高血压的AUC为0.936,明显优于DT(0.897)和LR(0.712)模型。DCA显示,XGBoost模型在训练集与测试集上的净获益高度接近(平均差异仅0.034 5),表明其在不同阈值概率下均具有稳定的临床决策价值。同时,模型在测试集上的判别性能(AUC=0.936)与训练集(AUC=0.971)相比下降0.035,灵敏度(92.31% vs. 90.11%)和F1值(0.867 5 vs. 0.926 6)亦保持较高水平,提示模型具备良好的泛化潜力。SHAP可解释性分析显示,维持性血液透析时间、平均超滤量、血磷、血清同型半胱氨酸、甲状旁腺激素(PTH)、血钙、透析液钠浓度、血清白蛋白、使用促红细胞生成素依次是影响模型预测的重要特征。结论: 基于维持性血液透析时间、平均超滤量、血磷、血清同型半胱氨酸、PTH、血钙、透析液钠浓度、血清白蛋白、使用促红细胞生成素初步构建的可解释性XGBoost模型展示了良好的区分能力,为未来开发透析中高血压风险预警工具提供了一种潜在的算法框架和特征选择思路。

Objective: To explore the predictive efficacy of different machine learning(ML) algorithm models for the risk of hypertension during hemodialysis in hemodialysis patients, aiming to provide decision support for early clinical identification of high-risk patients and implementation of individualized interventions. Methods: A retrospective cohort of 455 patients with end-stage renal disease undergoing hemodialysis in Kailuan General Hospital between August 2023 and August 2025 was selected. The entire dataset was randomly divided into a training set(n=319) and a test set(n=136) at a 7∶3 ratio using R software. Based on the occurrence of hypertension during hemodialysis, the training set was further divided into a hypertension group(n=130) and a non-hypertension group(n=189). Logistic regression(LR), decision tree(DT), and extreme gradient boosting(XGBoost) models for predicting hypertension during hemodialysis in hemodialysis patients were constructed using the training set. Model performance was evaluated on the test set using the area under the curve(AUC), sensitivity, specificity, and accuracy. Decision curve analysis(DCA) was employed to assess the clinical net benefit of the optimal model. Finally, Shapley additive explanations(SHAP) were applied to visually interpret the predictions of the optimal model. Results: The model performance evaluation results demonstrated that the XGBoost model, constructed using nine clinical variables, achieved an AUC of 0.936 for predicting hypertension during hemodialysis in hemodialysis patients on the test set, significantly outperforming the DT(0.897) and LR(0.712) models. The DCA revealed that the net benefit of the XGBoost model was highly consistent between the training and test sets(average difference of only 0.034 5), indicating stable clinical decision-making value across different threshold probabilities. Furthermore, the model's discriminative performance on the test set(AUC=0.936) decreased by only 0.035 compared to the training set(AUC=0.971). The sensitivity(92.31% vs. 90.11%) and F1 score(0.867 5 vs. 0.926 6) also remained at high levels, suggesting good generalization potential. SHAP interpretability analysis revealed that the duration of maintenance hemodialysis, average ultrafiltration volume, serum phosphorus, serum homocysteine, parathyroid hormone(PTH), serum calcium, dialysate sodium concentration, serum albumin, and erythropoietin use were, in descending order, the important features influencing model predictions. Conclusion: The preliminary interpretable XGBoost model, constructed using the duration of maintenance hemodialysis, average ultrafiltration volume, serum phosphorus, serum homocysteine, PTH, serum calcium, dialysate sodium concentration, serum albumin, and erythropoietin use, demonstrated strong discriminatory ability. This provides a potential algorithmic framework and feature selection approach for the future development of early warning tools for hypertension during hemodialysis risk.

参考文献：

[1] LIU Y,WEN H,BAI J,et al.Burden of diabetes and kidney disease attributable to non-optimal temperature from 1990 to 2019:a systematic analysis from the Global Burden of Disease Study 2019[J].Sci Total Environ,2022,838:156495.
[2] DEME S,JANAKIRAMAN B,ALAMER A,et al.Predictors of functional status and disability among patients living with chronic kidney diseases at St Paul's hospital millennium medical college,Ethiopia:findings from a cross-sectional study[J].BMC Nephrol,2024,25(1):343.
[3] 张沫,蒋仁莲,邵高海,等.我国近十年维持性血液透析患者抑郁患病率的Meta分析[J].东南大学学报(医学版),2024,43(3):355-363.
[4] TAKAHASHI T,KANAZAWA Y,NAKAO T.#2871 incremental hemodialysis is a cost-effective and patient-centered approach to renal replacement therapy for initiating outpatient dialysis[J].Nephrol Dial Transplant,2025,40(Supplement_3):gfaf116.0674.
[5] ALOSTAZ M,CORREA S,LUNDY G S,et al.Time of hemodialysis and risk of intradialytic hypotension and intradialytic hypertension in maintenance hemodialysis[J].J Hum Hypertens,2023,37(10):880-890.
[6] SAEED S A,ABDELWAHAAB SANAD K G,ALI EZZAT M,et al.Association between hyperparathyrodism and intradialytic hypertension in prevalent hemodialysis patients[J].QJM,2024,117(Supplement_2):hcae175.353.
[7] 姚盛华,王唯英,谢益女,等.血液透析患者并发高血压的影响因素分析[J].心电与循环,2025,44(2):193-197.
[8] ADEJUMO O A,EDEKI I R,OYEDEPO D S,et al.The prevalence and risk of mortality associated with intradialytic hypertension among patients with end-stage kidney disease on haemodialysis:a systematic review and meta-analysis[J].PLoS One,2024,19(6):e0304633.
[9] SAVAGE S A,SETH I,ANGUS Z G,et al.Advancements in microsurgery:a comprehensive systematic review of artificial intelligence applications[J].J Plast Reconstr Aesthetic Surg,2025,101:65-76.
[10] 田林,任绪泽,涂峥程.人工智能、机器学习和深度学习在医学诊断中的应用进展[J].现代医学,2024,52(9):1480-1484.
[11] WANG Y,AIVALIOTI E,STAMATELOPOULOS K,et al.Machine learning in cardiovascular risk assessment:Towards a precision medicine approach[J].Eur J Clin Investig,2025,55(S1):e70017.
[12] HASSAN J,SAEED S M,DEKA L,et al.Applications of machine learning(ML) and mathematical modeling(MM) in healthcare with special focus on cancer prognosis and anticancer therapy:current status and challenges[J].Pharmaceutics,2024,16(2):260.
[13] YUE S,LI S,HUANG X,et al.Machine learning for the prediction of acute kidney injury in patients with sepsis[J].J Transl Med,2022,20(1):215.
[14] NINAN J,NIKRAVANGOLSEFID N,TRUONG H H,et al.Prediction of intradialytic hypotension by machine learning:a systematic review[J].J Nephrol,2025,38(8):2077-2094.
[15] 谭敏,何文昌,柏晓鑫,等.基于Logistic回归构建维持性血液透析患者发生难治性高血压的预测模型研究[J].中国中西医结合肾病杂志,2024,25(12):1070-1073.
[16] WANG Y,ZHOU H,GUO Q,et al.Prediction model of intradialytic hypertension in hemodialysis patients based on machine learning[J].J Med Syst,2025,49(1):112.
[17] 上海慢性肾脏病早发现及规范化诊治与示范项目专家组.慢性肾脏病筛查诊断及防治指南[J].中国实用内科杂志,2017,37(1):28-34.
[18] 董一飞,董徽,许建忠,等.《中国高血压防治指南(2024年修订版)》更新要点解读[J].中华高血压杂志(中英文),2025,33(1):14-17.
[19] VAN BUREN P N,INRIG J K.Special situations:intradialytic hypertension/chronic hypertension and intradialytic hypotension[J].Semin Dial,2017,30(6):545-552.
[20] SINGH A T,WAIKAR S S,MC CAUSLAND F R.Association of different definitions of intradialytic hypertension with long-term mortality in hemodialysis[J].Hypertension,2022,79(4):855-862.
[21] RAIKOU V D,KYRIAKI D.The association between intradialytic hypertension and metabolic disorders in end stage renal disease[J].Int J Hypertens,2018,2018:1681056.
[22] 许清丽,晏现丽,刘丽,等.糖尿病肾病患者维持性血液透析发生难治性高血压的影响因素分析[J].中国中西医结合肾病杂志,2024,25(9):793-795.
[23] 钟霞,廖红霞,梁露叶.维持性血液透析患者透析后高血压的发生现状、危险因素及应对策略[J].现代医学与健康研究电子杂志,2024,8(7):88-90.
[24] 韩名力.终末期肾病血液透析患者并发高血压的相关影响分析[J].心血管病防治知识,2023,13(4):22-24.
[25] KUSAYAMA T,NAGAMORI Y,TAKEUCHI K,et al.Renal autonomic dynamics in hypertension:how can we evaluate sympathetic activity for renal denervation[J].Hypertens Res,2024,47(10):2685-2692.
[26] 黄瑶玲,张益民,温穗溱,等.糖尿病肾病患者透析过程中高血压的影响因素[J].国际泌尿系统杂志,2021,41(4):669-673.
[27] VILLA-ETCHEGOYEN C,LOMBARTE M,MATAMOROS N,et al.Mechanisms involved in the relationship between low calcium intake and high blood pressure[J].Nutrients,2019,11(5):1112.
[28] FENG X D,WANG S F,QIAO S Y,et al.Correlational analysis of resistant hypertension with diabetes mellitus,chronic kidney disease,and the interplay of sodium,calcium,magnesium,and phosphorus[J].Vasc Health Risk Manag,2025,21:217-228.
[29] IKRAM H,JAFFAR S R,HAYAT A,et al.Association of hypertension with raised serum homocysteine levels[J].Pak Armed Forces Med J,2022,72(3):1125-1129.
[30] 李霞,陈育青.影响维持性血液透析患者促红细胞生成素抵抗的因素分析[J].重庆医学,2022,51(9):1554-1557.
[31] DHARMARAJ R B,THANGAVEL K,INDIRAJITH V,et al.Serum albumin and uric acid levels in hypertensive patients:a cross-sectional analysis from central Tamil Nadu,south India[J].Cureus,2025,17(1):e76766.
[32] IATRIDI F,THEODORAKOPOULOU M P,EKART R,et al.The effect of low dialysate sodium concentration on ambulatory aortic blood pressure and arterial stiffness in patients with intradialytic hypertension:a randomized crossover study[J].Cureus,2025,17(1):e77079.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录