Using a Guided Machine Learning Ensemble Model to Predict Discharge Disposition following Meningioma Resection

Area of Science:

• Neurosurgery
• Machine Learning
• Predictive Analytics

Background:

• Machine learning (ML) algorithms offer significant potential for predicting patient outcomes.
• Accurate prediction of discharge disposition is crucial for effective patient management after meningioma resection.

Purpose of the Study:

• To pilot a novel machine learning approach for algorithm selection and ensemble model creation.
• To predict discharge disposition following meningioma resection using a proof-of-concept study.
• To compare the predictive performance of the developed ensemble model against traditional logistic regression.

Main Methods:

• Trained diverse ML algorithms on a single-institution meningioma patient database.
• Ranked algorithms by predictive power and created an ensemble model from top performers.
• Internally validated the ensemble model on unseen data and compared its performance to logistic regression.

Main Results:

• The ensemble model significantly outperformed logistic regression in predicting discharge disposition (AUC 0.78 vs. 0.71, p=0.01).
• Key variables influencing the model included tumor size, emergency department presentation, BMI, tumor location, and preoperative motor deficit.
• The ensemble model provided nuanced insights into the impact of individual variables.

Conclusions:

• A novel guided ML ensemble model was developed for predicting discharge destination after meningioma resection.
• This ML approach demonstrated superior predictive power compared to logistic regression.
• The methodology can be extended to predict various other patient outcomes, enhancing clinical decision-making.