Scholarlink

Objective <p>This study aimed to construct and validate an individualized prediction model for poor prognosis in elderly patients with Aneurysmal subarachnoid hemorrhage (aSAH) based on multimodal indicators such as clinical, imaging, and laboratory data.</p> Methods <p>We retrospectively enrolled 241 consecutive patients with aSAH from January 2017 to December 2020 as the training set. An independent temporal validation set included 104 consecutive patients from January to September 2024, with assignment strictly by chronological order. In the training set, univariate analysis identified candidate predictors of poor prognosis, which were refined via LASSO regression. Significant variables were then entered into multivariate logistic regression to define independent predictors. Using these predictors, we constructed random forest (RF), support vector machine (SVM), and k-nearest neighbor models (KNN). Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Model interpretability and variable contributions were analyzed with SHapley Additive exPlanations (SHAP).</p> Results <p>There were no statistically significant differences in the baseline clinical data between the training set and the validation set (all <i>P</i> &gt; 0.05). The results of multivariate Logistic regression analysis showed that World Federation of Neurosurgical Societies (WFNS) score, modified Fisher grade, intracerebral hematoma volume, maximum thickness of subarachnoid blood clots, C-reactive protein, and duration of symptomatic cerebral vasospasm were identified as independent risk factors for poor prognosis (all <i>P</i> &lt; 0.05). The performance evaluation of the machine-learning models showed that the SVM model had the best discrimination, with AUCs of 0.838 (95% CI: 0.764–0.912) and 0.791 (95% CI: 0.688–0.895) in the training set and the validation set, respectively. The calibration curve showed a high consistency between the predicted probability and the actual risk. DCA indicated that the model had clinical net benefits within a wide range of thresholds. SHAP analysis confirmed that C-reactive protein, maximum thickness of subarachnoid clot, and intracerebral hematoma volume were the most important contributing factors to the increased risk of poor prognosis.</p> Conclusion <p>This study successfully constructed and validated a prediction model for poor prognosis in elderly patients with aSAH based on multimodal indicators. The model has robust performance, clinically accessible indicators, and good interpretability, providing a valuable quantitative tool for dynamic risk stratification and implementation of stratified management.</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

Construction and verification of multimodal model for prognosis in elderly patients with aneurysmal subarachnoid hemorrhage

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月8日

查看全文
  • Wengao Zhang,
  • Gang Li,
  • Yue Gao,
  • Zhiguo Liu,
  • Baoxu Zhang,
  • Jinfeng Wang

摘要

Objective

This study aimed to construct and validate an individualized prediction model for poor prognosis in elderly patients with Aneurysmal subarachnoid hemorrhage (aSAH) based on multimodal indicators such as clinical, imaging, and laboratory data.

Methods

We retrospectively enrolled 241 consecutive patients with aSAH from January 2017 to December 2020 as the training set. An independent temporal validation set included 104 consecutive patients from January to September 2024, with assignment strictly by chronological order. In the training set, univariate analysis identified candidate predictors of poor prognosis, which were refined via LASSO regression. Significant variables were then entered into multivariate logistic regression to define independent predictors. Using these predictors, we constructed random forest (RF), support vector machine (SVM), and k-nearest neighbor models (KNN). Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Model interpretability and variable contributions were analyzed with SHapley Additive exPlanations (SHAP).

Results

There were no statistically significant differences in the baseline clinical data between the training set and the validation set (all P > 0.05). The results of multivariate Logistic regression analysis showed that World Federation of Neurosurgical Societies (WFNS) score, modified Fisher grade, intracerebral hematoma volume, maximum thickness of subarachnoid blood clots, C-reactive protein, and duration of symptomatic cerebral vasospasm were identified as independent risk factors for poor prognosis (all P < 0.05). The performance evaluation of the machine-learning models showed that the SVM model had the best discrimination, with AUCs of 0.838 (95% CI: 0.764–0.912) and 0.791 (95% CI: 0.688–0.895) in the training set and the validation set, respectively. The calibration curve showed a high consistency between the predicted probability and the actual risk. DCA indicated that the model had clinical net benefits within a wide range of thresholds. SHAP analysis confirmed that C-reactive protein, maximum thickness of subarachnoid clot, and intracerebral hematoma volume were the most important contributing factors to the increased risk of poor prognosis.

Conclusion

This study successfully constructed and validated a prediction model for poor prognosis in elderly patients with aSAH based on multimodal indicators. The model has robust performance, clinically accessible indicators, and good interpretability, providing a valuable quantitative tool for dynamic risk stratification and implementation of stratified management.