Predicting surgical decision-making in vestibular schwannoma using tree-based machine learning.

Area of Science:

• Neurosurgery
• Oncology
• Medical Informatics

Background:

• Vestibular schwannomas (VSs) are common cerebellopontine angle tumors with unpredictable growth.
• Treatment decisions involve balancing conservative management versus active intervention, often complicated by limited evidence and quality-of-life factors.

Purpose of the Study:

• To apply machine learning (ML) algorithms to predict treatment decisions for VS.
• To identify preoperative variables most influential in guiding active treatment versus MRI surveillance.

Main Methods:

• Tree-based ML algorithms (decision tree, random forest) were trained on patient data (2009-2021).
• Predictors included demographics, symptoms, and tumor characteristics; models were validated using cross-validation.

Main Results:

• Decision tree analysis achieved 85% accuracy predicting treatment based on tumor size (1.6 cm) and progressive symptoms.
• Optimized models with age at presentation reached 88% accuracy; random forest achieved 80% accuracy.
• Key predictive factors identified: maximum tumor dimension, patient age, Koos grade, and progressive symptoms.

Conclusions:

• Tree-based ML accurately predicts VS treatment decisions (80%-88% accuracy).
• Maximum tumor dimension, age, Koos grade, and symptom progression are critical decision drivers.
• ML provides an evidence-based approach to support surgical decision-making and patient counseling for VS.