Machine learning-based model to predict long-term tumor control and additional interventions following pituitary
Yuki Shinya1,2, Abdul Karim Ghaith1, Sukwoo Hong1,2
11Department of Neurologic Surgery, Mayo Clinic, Rochester, Minnesota.
Journal of Neurosurgery
|March 14, 2025
Summary
A machine learning model accurately predicts long-term remission in Cushing's disease after surgery. This tool helps personalize treatment by identifying key predictors like tumor size and BMI.
Area of Science:
- Endocrinology
- Neurosurgery
- Artificial Intelligence
Background:
- Cushing's disease (CD) management requires predicting long-term outcomes after endonasal transsphenoidal surgery (ETS).
- Accurate prediction of intervention-free survival (IFS) is crucial for patient management.
Purpose of the Study:
- To develop and validate a supervised machine learning (ML) model to predict long-term biochemical outcomes and IFS post-ETS in CD patients.
- To identify key predictors influencing long-term surgical success.
Main Methods:
- Retrospective review of 150 CD patients undergoing ETS (2013-2023).
- Development of decision tree and random forest ML models to predict IFS using baseline, surgical, and endocrine data.
- 80/20 train-test split for model validation.
Main Results:
- The decision tree ML model achieved 91% accuracy in predicting IFS on an unseen test dataset.
- Tumor size on MRI, Knosp grade, patient age, and BMI were identified as significant predictors of long-term IFS.
- Intervention-free survival rates were 83% at 3 years and 78% at 5 years post-ETS.
Conclusions:
- ML models can accurately predict long-term remission in Cushing's disease after ETS.
- Prognosis is influenced by factors including tumor size, Knosp grade, age, and BMI.
- The developed decision tree model can aid in patient stratification and personalized treatment planning for CD.
