Machine learning-based clinical outcome prediction in surgery for acromegaly
Olivier Zanier1, Matteo Zoli2,3, Victor E Staartjes1
1Machine Intelligence in Clinical Neuroscience (MICN) Laboratory, Department of Neurosurgery, Clinical Neuroscience Center, University Hospital Zurich, University of Zurich, Zurich, Switzerland.
Endocrine
|October 13, 2021
Summary
Machine learning models can predict outcomes like gross total resection (GTR) and biochemical remission (BR) after acromegaly surgery. While challenging, these models show promise for tailoring surgical treatments.
Area of Science:
- Neurosurgery
- Endocrinology
- Machine Learning in Medicine
Background:
- Transsphenoidal surgery is a key treatment for acromegaly.
- Predicting surgical outcomes like biochemical remission (BR), gross total resection (GTR), and cerebrospinal fluid (CSF) leaks is crucial for patient management.
Purpose of the Study:
- To develop and externally validate machine learning models for predicting GTR, BR, and intraoperative CSF leaks after transsphenoidal surgery for acromegaly.
Main Methods:
- Machine learning models were developed using data from a Bologna, Italy registry.
- External validation was performed on a separate registry from Zurich, Switzerland.
- Input features included gender, age, prior surgery, and Hardy and Knosp classifications.
Main Results:
- The derivation cohort included 307 patients; external validation involved 46 patients.
- External validation showed AUCs of 0.75 for GTR, 0.63 for BR, and 0.77 for CSF leaks.
- Prior surgery was key for GTR prediction, while age and Hardy grading were important for BR and CSF leak predictions.
Conclusions:
- Predicting GTR, BR, and CSF leaks remains difficult but machine learning shows potential.
- The study demonstrates the feasibility of developing and validating prediction models for acromegaly surgery outcomes.
- This work lays the foundation for a multicenter model with improved generalizability.
