Predicting Survival of Patients with Spinal Ependymoma Using Machine Learning Algorithms with the SEER Database

Area of Science:

• Neuro-oncology
• Clinical data analysis
• Machine learning in medicine

Background:

• Spinal ependymomas are primary tumors of the central nervous system.
• Understanding factors influencing overall survival (OS) is crucial for patient management.
• Predictive models can aid in clinical decision-making and prognosis.

Purpose of the Study:

• To identify clinical and demographic factors affecting OS in spinal ependymoma.
• To develop and evaluate machine learning (ML) algorithms for OS prediction.
• To compare ML performance against traditional statistical methods.

Main Methods:

• Utilized data from the Surveillance, Epidemiology, and End Results (SEER) registry (1973-2014).
• Performed statistical analyses including Kaplan-Meier and Cox regression.
• Implemented and assessed ML algorithms for predicting 5- and 10-year OS.

Main Results:

• Independent OS predictors identified: age ≥65, histologic subtype, metastasis, multiple lesions, surgery, radiation, and gross total resection (GTR).
• ML model achieved AUC of 0.74 (5-year) and 0.81 (10-year) for OS prediction.
• ML outperformed stepwise logistic regression (AUC 0.71/0.75).

Conclusions:

• Therapeutic interventions like surgery and GTR significantly improve OS in spinal ependymoma.
• ML techniques demonstrate robust performance in predicting OS compared to statistical models.
• Dataset heterogeneity and missing values present challenges for predictive modeling.