Prehospital stroke-scale machine-learning model predicts the need for surgical intervention

Area of Science:

• Neurology
• Medical Technology
• Data Science

Background:

• Prehospital diagnosis scales for stroke are developing globally.
• Machine learning models offer potential for advanced stroke assessment.

Purpose of the Study:

• To evaluate a novel scale predicting surgical intervention needs in stroke patients.
• To assess the scale's accuracy across different stroke types, including subarachnoid and intracerebral hemorrhage.

Main Methods:

• A multicenter retrospective study analyzed 23 prehospital items (vitals, neurological symptoms) in stroke patients.
• An eXtreme Gradient Boosting (XGBoost) model was developed for binary classification of surgical intervention need.
• Data from 1143 patients were split into training (70%) and testing (30%) cohorts.

Main Results:

• The XGBoost model achieved high accuracy in predicting surgical intervention (AUC 0.802).
• Key predictors included level of consciousness, vital signs, sudden headache, and speech abnormalities.
• The model demonstrated strong sensitivity (0.748) and specificity (0.853) in the test cohort.

Conclusions:

• The developed algorithm effectively predicts the need for surgical intervention in prehospital stroke assessment.
• This tool can significantly enhance prehospital stroke management and contribute to better patient outcomes.