Machine Learning–Driven Risk Prediction of Delayed Cerebral Ischemia After Aneurysmal Subarachnoid Hemorrhage Using Peripheral Inflammatory Markers

Yuanyuan Liu¹Chengchen Li¹Honglin Wang^2*

• ¹Chengdu Women and Children's Central Hospital, Chengdu, China
• ²Dazhou Central Hospital, DaZhou, China

Background: Delayed cerebral ischemia (DCI) remains a leading cause of secondary neurological deterioration and mortality after aneurysmal subarachnoid hemorrhage (aSAH). Accumulating evidence highlights the pivotal role of systemic inflammation in the pathogenesis of DCI, with peripheral inflammatory markers showing potential as early indicators. However, the predictive performance of individual biomarkers is limited. By leveraging machine learning (ML) techniques, it is possible to integrate heterogeneous inflammatory signals and model complex nonlinear relationships to improve individualized risk prediction. Methods and materials: We conducted a retrospective analysis of 562 aSAH patients admitted to a single tertiary center. Clinical, radiographic, and laboratory data—including peripheral inflammatory indices—were extracted from electronic medical records. The Boruta algorithm was applied for feature selection. Six ML models were developed and compared: logistic regression, neural network, random forest, support vector machine, gradient boosting machine (GBM), and extreme gradient boosting (XGBoost). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, F1 score, calibration curves, and decision curve analysis (DCA). Results: Among the six models, the neural network demonstrated the best balance between discrimination and calibration, with an AUC of 0.826 in the training cohort and 0.808 in the internal testing cohort. Eight predictors were included in the final model: Glasgow Coma Scale (GCS), Hunt-Hess grade, modified Fisher score, prognostic nutritional index (PNI), neutrophil-to-albumin ratio (NAR), neutrophil-to-lymphocyte platelet ratio (NLPR), C-reactive protein-to-lymphocyte ratio (CLR), and procalcitonin. SHapley Additive exPlanations (SHAP) analysis revealed Hunt-Hess grade and procalcitonin as top contributors. Conclusions: This study proposes a machine learning–based risk prediction tool for DCI after aSAH, built from routinely available inflammatory and clinical variables. The model demonstrated strong discriminative and calibration performance and provides a clinically interpretable, preoperative decision-support tool. Prospective multicenter validation is warranted to assess generalizability and facilitate clinical translation.