Scholarlink

Background <p>External ventricular drain (EVD) weaning trials assess the ability of patients with nontraumatic subarachnoid hemorrhage (SAH) to maintain normal intracranial pressure (ICP) without external drainage. Early weaning outcome prediction may optimize clinical management. Existing studies rely on intermittent ICP measurements or static electronic health record data, failing to capture dynamic ICP information. We hypothesize that similar mean ICP produces similar pulse morphology under homeostatic conditions, and acute ICP dynamics changes alter this pattern.</p> Methods <p>We retrospectively analyzed 588 patients with SAH with 413 successful and 239 failed EVD weaning attempts. Dominant ICP pulses were extracted, and morphological similarity was quantified between pulse pairs. A reference pulse similarity distribution was established from a random 50% of successful attempts. Kullback–Leibler divergence measured deviations from this reference distribution for remaining attempts. Three machine learning models compared performance using pulse similarity alone versus combined with key clinical features (age, sex, World Federation of Neurological Societies scale, modified Fisher scale).</p> Results <p>All models showed progressive performance improvement over time in predicting clamp trial result from the time of EVD clamping. Support vector machine achieved superior performance with pulse features alone, reaching F1 score of 0.77 (± 0.07, CI 0.69–0.85), AUC of 0.86 (± 0.05, CI 0.79–0.93), and Brier score of 0.15 (± 0.01, CI 0.13–0.17) at 24&#xa0;h. Clinical features provided modest improvements at early hours (5–18% AUC increase) with diminishing benefits later. Decision curve analysis revealed incorporating clinical features enhanced clinical utility from 10&#xa0;h onward, even with modest discriminative ability, while models achieved marked clinical utility at 16–24&#xa0;h.</p> Conclusions <p>This study demonstrates that ICP pulse morphological similarity can predict EVD weaning outcomes and reveals the emergence of clinical value beginning 10&#xa0;h after trial initiation, potentially enabling optimizing patient care in SAH management complicated with obstructive hydrocephalus.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Early Prediction of External Ventricular Drain Weaning Trial Outcome in Patients with SAH Based on ICP Pulse Morphological Similarity: A Single-Center Retrospective Cohort Analysis

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Kejun Dong,
  • Ofer Sadan,
  • Paul Horton,
  • Fahimeh Mohagheghian,
  • Delgersuren Bold,
  • Xiao Hu

摘要

Background

External ventricular drain (EVD) weaning trials assess the ability of patients with nontraumatic subarachnoid hemorrhage (SAH) to maintain normal intracranial pressure (ICP) without external drainage. Early weaning outcome prediction may optimize clinical management. Existing studies rely on intermittent ICP measurements or static electronic health record data, failing to capture dynamic ICP information. We hypothesize that similar mean ICP produces similar pulse morphology under homeostatic conditions, and acute ICP dynamics changes alter this pattern.

Methods

We retrospectively analyzed 588 patients with SAH with 413 successful and 239 failed EVD weaning attempts. Dominant ICP pulses were extracted, and morphological similarity was quantified between pulse pairs. A reference pulse similarity distribution was established from a random 50% of successful attempts. Kullback–Leibler divergence measured deviations from this reference distribution for remaining attempts. Three machine learning models compared performance using pulse similarity alone versus combined with key clinical features (age, sex, World Federation of Neurological Societies scale, modified Fisher scale).

Results

All models showed progressive performance improvement over time in predicting clamp trial result from the time of EVD clamping. Support vector machine achieved superior performance with pulse features alone, reaching F1 score of 0.77 (± 0.07, CI 0.69–0.85), AUC of 0.86 (± 0.05, CI 0.79–0.93), and Brier score of 0.15 (± 0.01, CI 0.13–0.17) at 24 h. Clinical features provided modest improvements at early hours (5–18% AUC increase) with diminishing benefits later. Decision curve analysis revealed incorporating clinical features enhanced clinical utility from 10 h onward, even with modest discriminative ability, while models achieved marked clinical utility at 16–24 h.

Conclusions

This study demonstrates that ICP pulse morphological similarity can predict EVD weaning outcomes and reveals the emergence of clinical value beginning 10 h after trial initiation, potentially enabling optimizing patient care in SAH management complicated with obstructive hydrocephalus.