文献速递脑血管病外科小儿神经外科与脑积水神经重症

机器学习预测动脉瘤性蛛网膜下腔出血后慢性正常压力脑积水行脑室腹腔分流术的认知改善

Machine Learning in Predicting the Cognitive Improvement of Ventriculoperitoneal Shunt for Chronic Normal Pressure Hydrocephalus After Aneurysmal Subarachnoid Hemorrhage.
PubMed 文献 · World Neurosurg · 2025-03-11

本研究纳入75例动脉瘤性蛛网膜下腔出血后慢性正常压力脑积水行脑室腹腔分流术的患者,采用8种监督机器学习算法构建预测模型评估分流获益。48例(64%)获益,27例(36%)未获益;极端梯度提升模型曲线下面积最高为0.946,并据此构建了列线图。

要点

  • 动脉瘤性蛛网膜下腔出血后慢性正常压力脑积水患者行脑室腹腔分流术,48例(64%)获益,27例(36%)未获益。
  • 获益组病程更长、脑脊液压力更高、脑脊液红细胞和白细胞计数更低、改良Fisher分级和Hunt-Hess分级更低。
  • 极端梯度提升模型预测性能最佳,曲线下面积为0.946,并构建了列线图。

核心结论

  • 动脉瘤性蛛网膜下腔出血后慢性正常压力脑积水患者行脑室腹腔分流术的获益与病程、脑脊液压力、脑脊液红细胞/白细胞计数、血白细胞计数、改良Fisher分级、Hunt-Hess分级及术前改良Rankin量表评分相关。
  • 极端梯度提升模型预测性能最佳,曲线下面积为0.946。

研究设计与样本

  • 纳入动脉瘤性蛛网膜下腔出血后慢性正常压力脑积水且接受脑室腹腔分流术的患者,共75例(男39例,女36例)。
  • 分析手术前后改良Rankin量表变化以评估分流获益,采用最小绝对收缩和选择算子筛选变量,并用8种监督机器学习算法构建预测模型。

主要发现

  • 75例患者中,48例(64%)从脑室腹腔分流术获益,27例(36%)未获益。
  • 获益组较未获益组病程更长、脑脊液压力更高、脑脊液红细胞和白细胞计数更低、改良Fisher分级和Hunt-Hess分级更低。
  • 单因素逻辑回归分析显示,病程、脑脊液压力、脑脊液红细胞/白细胞计数、血白细胞计数、改良Fisher分级、Hunt-Hess分级及术前改良Rankin量表评分与分流术良好结局相关。
  • 极端梯度提升模型曲线下面积最高为0.946,残差误差最低,据此构建的列线图表现满意。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: Chronic normal pressure hydrocephalus (CNPH) is a recognized sequela of aneurysmal subarachnoid hemorrhage (ASAH). Ventriculoperitoneal shunt (VPS) is a conventional treatment for hydrocephalus, though its effectiveness for CNPH post-ASAH remains unclear. METHODS: We included ASAH patients with CNPH who underwent VPS surgery. Changes in the modified Rankin Scale (mRS) before and after surgery were analyzed to evaluate VPS benefits. The least absolute shrinkage and selection operator identified relevant variables and predictive models were constructed using 8 supervised machine learning algorithms to assess VPS benefit. RESULTS: Among 75 patients (39 males and 36 females), 48 (64%) benefited from VPS, while 27 (36%) did not. The beneficial group showed a longer disease course, higher cerebrospinal fluid (CSF) pressure, lower red and white blood cell counts in CSF, and lower modified Fisher (MF) and Hunt-Hess (HH) grades compared to the nonbeneficial group. Univariate logistic regression analysis indicated that disease course, CSF pressure, red blood cell/white blood cell (WBC) counts in CSF, WBC count in blood, MF grade, HH grade, and preoperative mRS were associated with favorable VPS outcomes. The extreme gradient boosting (XGB) model demonstrated the highest area under the curve of 0.946 and lowest residual error. A nomogram was subsequently developed and demonstrated a satisfactory performance. CONCLUSIONS: VPS benefits in CNPH patients after ASAH were associated with disease course, CSF pressure, red blood cell/WBC counts in CSF, WBC count in blood, MF and HH grades, and preoperative mRS. The XGB model demonstrated optimal predictive performance, with an area under the curve of 0.946.

来源:World Neurosurg,查看原文。版权归原作者所有。

引用

  1. Zhu Y, Xue L, Zhao F, et al. Machine Learning in Predicting the Cognitive Improvement of Ventriculoperitoneal Shunt for Chronic Normal Pressure Hydrocephalus After Aneurysmal Subarachnoid Hemorrhage. World Neurosurg. 2025. PMID: 39952404. doi:10.1016/j.wneu.2025.123771

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。