中国血液净化

中国血液净化 ›› 2025, Vol. 24 ›› Issue (06): 474-478.doi: 10.3969/j.issn.1671-4091.2025.06.007

• 临床研究 • 上一篇    下一篇

血液透析相关性头痛预测模型的构建和验证

邹 涵   刘 霞   胡晓霞    李 洁    张艳芳    李子涵   

  1. 730000 兰州，兰州大学第二医院(第二临床医学院)1肾内科 2特需内科 3神经内科 4疼痛科
  • 收稿日期:2024-12-05 修回日期:2025-02-08 出版日期:2025-06-12 发布日期:2025-06-12
  • 通讯作者: 邹涵 E-mail:906799145@qq.com
  • 基金资助:
    甘肃省自然科学项目

Construction and verification of hemodialysis-related headache prediction model base on Logistic-Nomogram

ZOU Han,LIU Xia, HU Xiao-xia, LI Jie, ZHANG Yan-fang, LI Zi-han   

  1. Department of Nephrology, 2Department of Special Internal Medicine, 3Department of Neurology, and 4Department of Pain, Lanzhou University Second Hospital (Second Clinical School), Lanzhou 730000, China
  • Received:2024-12-05 Revised:2025-02-08 Online:2025-06-12 Published:2025-06-12
  • Contact: 730000 兰州，兰州大学第二医院(第二临床医学院)2特需内科 E-mail:906799145@qq.com

摘要： 目的 探讨血液透析相关性头痛(hemodialysis-related headache，HRH)的危险因素，基于Logistic-Nomogram构建HRH的预测模型并进行验证。 方法  回顾性分析2021年3月—2024年3月在兰州大学第二医院进行血液透析患者的临床资料，将2021年3月—2023年5月、2023年6月—2024年3月的血液透析患者分别纳入训练集、验证集。采用Logistic回归模型分析HRH发生的影响因素；采用R语言中的rms包构建Nomogram预测模型；采用R语言中的calibrate函数验证预测曲线与理想曲线的贴合度，采用ROC曲线分析基于多因素Logistics回归结果构建的预测模型对HRH发生风险的预测价值。 结果 共纳入327例患者，其中训练集229例，验证集98例。Logistics回归分析显示：透析前收缩压(systolic blood pressure，SBP)、透析前舒张压(diastolic blood pressure，DBP)、血清钠升高是患者发生HRH的独立危险因素(OR分别为1.124、1.128、1.119，95% CI分别为1.051～1.203、1.066～1.194、1.076～1.338，P分别为0.001、0.001和＜0.001)，血小板计数(PLT)升高是发生HRH的保护因素(OR=0.932，95% CI：0.895～0.971，P=0.001)。以SBP、DBP、血清钠、PLT为变量，构建HRH发生风险的预测模型，模型在训练集Dxy=0.831，C指数=0.916，在验证集Dxy=0.804，C指数=0.902，模型预测效能较好。训练集预测曲线的平均绝对误差为0.017，验证集预测曲线的平均绝对误差为0.029，训练集与验证集的预测曲线均接近对角线，预测曲线与实际理想曲线拟合度良好。ROC曲线分析显示模型在训练集的AUC为0.916(95% CI：0.872～0.948)，最大约登指数为0.720，敏感性81.13%，特异性90.91%；模型在验证集的AUC为0.903(95% CI：0.826～0.954)，最大约登指数为0.756，敏感性92%，特异性83.56%。 结论  透析前SBP、透析前DBP、血清钠升高是发生HRH的独立危险因素，PLT升高是保护因素。以SBP、DBP、血清钠、PLT为变量构建的预测模型对血液透析患者HRH的发生风险具有较高的预测价值。

关键词: 血液透析相关性头痛, 影响因素, 列线图, 预测模型

Abstract: Objective  To investigate risk factors for hemodialysis-related headache (HRH) and develop and validate a predictive model using logistic-nomogram analysis. Methods  Clinical data of hemodialysis patients at the Second Hospital of Lanzhou University from March 2021 to March 2024 were retrospectively analyzed. Patients treated from March 2021 to May 2023 were assigned to the training set (n=229), and those from June 2023 to March 2024 were included in the validation set (n=98). The influencing factors of HRH were analyzed by Logistic regression models. A nomogram model was constructed using the rms package in R, with calibration evaluated via the calibrate function. Receiver operating characteristic (ROC) curves assessed the model’s predictive performance. Results  There was no significant difference in clinical data between the training set and the validation set (P＞0.05). Logistic regression analysis showed that elevated pre-dialysis systolic blood pressure (SBP, OR =1.124, 95% CI: 1.051～1.203, P=0.001), diastolic blood pressure (DBP, OR=1.128, 95% CI: 1.066～1.194, P=0.001), and serum sodium (OR=1.119, 95% CI:1.076～1.338, P＜0.001) were independent risk factors for HRH in hemodialysis patients, while higher platelet count (PLT) was a protective factor for HRH (OR=0.932, 95% CI:0.895～0.971, P=0.001). The prediction model of HRH risk in hemodialysis patients was constructed with SBP, DBP, serum sodium and PLT as variables. The nomogram model demonstrated strong predictive performance, Dxy=0.831, C-index=0.916, mean absolute error (MAE) =0.017 in the training set, and Dxy=0.804, C-index=0.902, MAE=0.029 in the validation set. Calibration curves closely aligned with ideal curves in both sets.  ROC curve analysis showed that the area under the curve (AUC) =0.916 (95% CI: 0.872～0.948), Youden index=0.720, sensitivity=81.13%, specificity=90.91% in the training set, and AUC=0.903 (95% CI: 0.826～0.954), Youden index=0.756, sensitivity=92%, specificity= 83.56% in the validation set.  Conclusion  Elevated pre-dialysis SBP, DBP, and serum sodium are independent risk factors for HRH, while higher PLT is protective. The nomogram model based on these variables provides robust predictive value for HRH risk in hemodialysis patients.

Key words: Hemodialysis-related headache, Influencing factors, Nomogram, Prediction model

中图分类号: