Predicting survival of spinal ependymoma patients using machine learning algorithms with SEER database.

Predicting survival of spinal ependymoma patients using machine learning algorithms with SEER database.

World Neurosurg. 2018 Dec 28;:

Authors: Ryu SM, Lee SH, Kim ES, Eoh W

Abstract
OBJECTIVE: This study was conducted to understand the clinical and demographic factors influencing the overall survival (OS) of spinal ependymoma patients 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 machine learning algorithms to predict the OS of spinal ependymoma patients.
RESULTS: In the multivariate analysis model, age ≥ 65 years, histological 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 although the dataset was heterogeneous and complex with numerous missing values.

PMID: 30597279 [PubMed - as supplied by publisher]

from A via a.sfakia on Inoreader http://bit.ly/2AmUi4k