Predicting Clinical Outcomes of Large Vessel Occlusion Before Mechanical Thrombectomy Using Machine Learning

Area of Science:

• Neurology
• Artificial Intelligence in Medicine
• Medical Informatics

Background:

• Acute ischemic stroke with large vessel occlusion (LVO) has a complex clinical course influenced by multiple prognostic factors.
• Predicting long-term outcomes in LVO patients before endovascular treatment is crucial for clinical decision-making.

Purpose of the Study:

• To develop and validate machine learning models for predicting long-term clinical outcomes in LVO patients prior to endovascular therapy.
• To compare the predictive performance of machine learning models against established pretreatment scoring methods.

Main Methods:

• Utilized logistic regression and machine learning algorithms (regularized logistic regression, linear support vector machine, random forest) on a derivation cohort (387 LVO patients) and an external validation cohort (115 LVO patients with anterior circulation).
• Predictors included standard and multiple pretreatment clinical variables.
• Compared model performance using the area under the receiver operating characteristic curve (AUC) against five previously reported scoring methods.

Main Results:

• The random forest model demonstrated superior predictive accuracy in both derivation (AUC=0.85±0.07) and validation cohorts (AUC=0.87±0.01) compared to logistic regression and existing scoring systems.
• Machine learning algorithms, particularly random forest, significantly outperformed standard statistical models and the best-performing prior scoring methods (e.g., Pittsburgh Response to Endovascular Therapy).

Conclusions:

• Machine learning models incorporating multiple pretreatment clinical variables offer enhanced accuracy in predicting clinical outcomes for patients with anterior circulation LVO undergoing mechanical thrombectomy.
• These AI-driven predictive tools surpass the performance of previously developed pretreatment scoring methods, paving the way for more precise prognostic assessments in LVO stroke management.