Machine Learning Predicts Cerebral Vasospasm in Subarachnoid Hemorrhage Patients.

Area of Science:

• Neurology
• Intensive Care Medicine
• Machine Learning in Healthcare

Background:

• Cerebral vasospasm (CV) is a critical complication post-subarachnoid hemorrhage (SAH), leading to delayed cerebral ischemia.
• Current management involves prolonged, resource-intensive intensive care unit (ICU) monitoring for SAH patients.
• Predicting CV requiring verapamil (CVRV) is crucial for timely intervention and resource optimization.

Approach:

• Utilized a multi-center dataset of SAH patients from UCLA and VUMC.
• Extracted 172 ICU variables for each patient.
• Trained a Light Gradient Boosting Machine (LightGBM) model with cross-validation to predict CVRV.

Key Points:

• The LightGBM model achieved an AUC of 0.88, predicting CVRV over a week in advance.
• Key predictive variables included minimum leukocyte count, maximum platelet count, and maximum intracranial pressure.
• Models demonstrated high accuracy in predicting CVRV timing (within or after three days).

Conclusions:

• Developed an accurate and early machine learning predictor for CVRV in SAH patients.
• The model's performance was validated across two independent institutions.
• This tool offers a significant advancement for optimizing clinical management and resource allocation in ICUs for SAH patients.