Ensemble machine learning models show best discriminative performance for predicting DCI after aneurysmal SAHMachine learning models show promise for predicting brain complications after aneurysm rupture

This systematic review and meta-analysis synthesized evidence from 29 studies involving approximately 10,000 patients with aneurysmal subarachnoid hemorrhage to evaluate machine learning models for predicting delayed cerebral ischemia. The analysis compared various algorithmic approaches, with ensemble methods (Random Forest, XGBoost) showing the best discriminative performance, achieving median AUCs of 0.80-0.85. Logistic regression was the most commonly used and interpretable model, while deep learning models demonstrated variable performance and greater overfitting. Sensitivity and specificity metrics varied across different models and patient cohorts, with ensemble approaches generally providing the best balance between these predictive metrics.

No safety or tolerability data were reported for the machine learning models themselves, as this analysis focused on predictive performance rather than clinical intervention outcomes. The primary limitation identified was the scarcity of external validation across studies, meaning most models were tested on the same datasets used for their development rather than independent patient populations.

From a practice perspective, machine learning models—particularly ensemble approaches—show promise for improving DCI prediction after SAH, potentially offering better discrimination than traditional statistical methods. However, the evidence represents a synthesis of observational prediction studies rather than clinical trials, and the models have not been compared against standard clinical prediction rules or decision-making processes. The findings should be interpreted as demonstrating technical feasibility rather than proven clinical utility, with significant validation work needed before any clinical application.