125 Leveraging Machine Learning to Develop a Novel, Intuitive, and Interpretable Risk Assessment Score for Predicting Local Tumor Progression After Stereotactic Radiosurgery for Spinal Metastases

View at publisher Summarise with VIRAT
Authors: Suchet Taori, Samuel Adida, Shovan Bhatia, Michael Ryan Kann, James C. Bayley, Pascal N. Zinn, Steven A. Burton, John Flickinger, Roberta Kay Sefcik, Peter Carlos Gerszten
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: Local tumor progression (LTP) after stereotactic radiosurgery (SRS) for spinal metastases can result in the development of significant pain and neurologic decline, yet predictive tools to stratify LTP risk and guide patient selection and follow-up are currently lacking. METHODS: A prospectively maintained database of 787 SRS treatments for spinal metastases from 2001-2024 was analyzed. Twenty-two clinical, radiographic, and treatment-related variables were evaluated to predict LTP. Five high-fidelity supervised machine learning classifiers (logistic regression, random-forest, gradient-boosted decision tree, Gaussian naïve Bayes, and support vector machine models) were optimized using hyperparameter tuning and trained with five-fold cross-validation (80% training, 20% testing). Models were evaluated using area under the receiver operating characteristic curve and F1 scores. SHapley Additive exPlanations (SHAP) were used to rank feature importance. The most predictive model was then used to construct a normalized point-based scoring system. RESULTS: At a median follow-up of 9 months (IQR: 2-23), radiographically-proven LTP occurred in 111 (14.1%) cases. the final model incorporated five weighted features to create a 0-15-point risk score: Spinal instability Neoplastic Score (SINS) >6 (4 points), lytic lesion (4 points), spinal misalignment (3 points), no prior chemotherapy (2 points), and polymetastatic disease (2 points). the model concordance index was 0.808. Crude LTP incidences in low-risk (LTP score: 0-5), intermediate-risk (LTP score: 6-8), and high-risk (LTP score: 9-15) groups were 2.4%, 10.2%, and 35.5%, respectively. Stratified survival analyses demonstrated significant LTP differences across all risk groups (log-rank and Gray’s test, p

Fields of Study & Subjects

Management of metastatic bone disease Brain Metastases and Treatment Spinal Fractures and Fixation Techniques

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
18–18
Publication date
26 March 2026
Language
EN