Machine learning as a clinical decision support tool for patients with acromegaly
Cem Sulu1, Ayyüce Begüm Bektaş2, Serdar Şahin1
1Department of Internal Medicine, Division of Endocrinology, Metabolism, and Diabetes, Cerrahpasa Medical School, Istanbul University-Cerrahpaşa, Kocamustafapaşa Street No:53, 34098 Fatih, Istanbul, Turkey.
Pituitary
|April 18, 2022
Summary
Machine learning models accurately predict acromegaly remission and somatostatin receptor ligand resistance. Key predictors include preoperative growth hormone, age, and tumor size, aiding treatment strategies.
Area of Science:
- Endocrinology
- Medical Informatics
- Machine Learning
Background:
- Acromegaly, a rare endocrine disorder, results from excess growth hormone (GH).
- Predicting treatment outcomes like remission and somatostatin receptor ligand (SRL) resistance is crucial for patient management.
- Current prediction methods may lack precision, necessitating advanced analytical tools.
Purpose of the Study:
- To develop and validate machine learning (ML) models for predicting postoperative remission, long-term remission, and SRL resistance in acromegaly patients.
- To identify key clinical features influencing these treatment outcomes using explainable AI techniques.
Main Methods:
- Utilized area under the receiver operating characteristics (AUROC) curves to evaluate model performance.
- Employed Shapley Additive explanations (SHAP) to interpret feature importance and model predictions.
- Developed extreme gradient boosting models for classification tasks.
Main Results:
- ML models achieved significant predictive performance: AUROC of 0.728 for early remission, 0.879 for last visit remission, and 0.753 for SRL resistance.
- Preoperative GH, age, and tumor size were critical for early remission prediction.
- Tumor size and SRL resistance predicted remission at last visit.
- Postoperative IGF1, GH levels, and adenoma subtype predicted SRL resistance.
Conclusions:
- Machine learning models demonstrate substantial potential as valuable tools for predicting remission and SRL resistance in acromegaly.
- These models can aid clinicians in tailoring treatment strategies and managing patient expectations.
