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

OBJECTIVE: This study was conducted to understand the clinical and demographic factors influencing the overall survival (OS) of patients with spinal ependymoma and to predict the OS with machine learning (ML) algorithms. METHODS: We compiled spinal ependymoma cases diagnosed between 1973 and 2014 from the Surveillance, Epidemiology, and End Results (SEER) registry. To identify the factors influencing survival, statistical analyses were performed using the Kaplan-Meier method and Cox proportional hazards regression model. In addition, we implemented ML algorithms to predict the OS of patients with spinal ependymoma. RESULTS: In the multivariate analysis model, age >= 65 years, histologic subtype, extraneural metastasis, multiple lesions, surgery, radiation therapy, and gross total resection (GTR) were found to be independent predictors for OS. Our ML model achieved an area under the receiver operating characteristic curve (AUC) of 0.74 (95% confidence interval [CI], 0.72-0.75) for predicting a 5-year OS of spinal ependymoma and an AUC of 0.81 (95% CI, 0.80-0.83) for predicting a 10-year OS. The stepwise logistic regression model showed poorer performance by an AUC of 0.71 (95% CI, 0.70-0.72) for predicting a 5-year OS and an AUC of 0.75 (95% CI, 0.73-0.77) for predicting a 10-year OS. CONCLUSIONS: With SEER data, we reaffirmed that therapeutic factors, such as surgery and GTR, were associated with improved OS. Compared with statistical methods, ML techniques showed satisfactory results in predicting OS; however, the dataset was heterogeneous and complex with numerous missing values.