The State of Machine Learning in Outcomes Prediction of Transsphenoidal Surgery: A Systematic Review
Darrion B Yang1, Alexander D Smith1, Emily J Smith1
1Carle Illinois College of Medicine, University of Illinois Urbana Champaign, Champaign, Illinois, United States.
Journal of Neurological Surgery. Part B, Skull Base
|October 19, 2023
Summary
Machine learning (ML) algorithms show promise in predicting transsphenoidal surgery outcomes like complications and recurrence. These advanced models, particularly ensemble methods and neural networks, offer clinical utility and can aid surgical decision-making.
Area of Science:
- Neurosurgery
- Medical Informatics
- Artificial Intelligence
Background:
- Transsphenoidal surgery is a critical procedure for various pituitary and skull base pathologies.
- Predicting postoperative outcomes remains a challenge, impacting patient management and recovery.
- The integration of advanced computational tools is essential for improving surgical precision and patient care.
Purpose of the Study:
- To systematically review and assess the application of machine learning (ML) algorithms in predicting postoperative outcomes following transsphenoidal surgery.
- To evaluate the performance and clinical utility of ML models in forecasting complications, recurrence, and mortality.
- To identify key features utilized by ML algorithms for outcome prediction in this surgical context.
Main Methods:
- A systematic review adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines was conducted.
- Searches of Scopus, PubMed, and Web of Science databases identified relevant studies published before May 12, 2021.
- Data extraction included study characteristics, ML model performance metrics (sensitivity, specificity, AUC), and important predictive features.
Main Results:
- Thirteen studies involving 5,048 patients were analyzed, with a focus on predicting outcomes for adenomas, acromegaly, and Cushing's disease.
- All included studies reported an Area Under the Curve (AUC) greater than 0.7, indicating significant clinical utility for the ML models.
- Ensemble algorithms and neural networks frequently demonstrated superior performance; biochemical and preoperative factors were commonly identified as crucial predictors.
Conclusions:
- Machine learning algorithms possess substantial potential for accurately predicting postoperative outcomes in transsphenoidal surgery.
- These predictive capabilities can significantly enhance clinical decision-making and ultimately improve patient care and surgical planning.
- Addressing the challenge of model interpretability through methods like LIME and Shapley values is crucial for broader clinical adoption.
