Using machine learning techniques predicts prognosis of patients with Ewing sarcoma

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Ewing sarcoma is a prevalent malignant bone tumor in pediatric and adolescent populations.
• Predictive models for Ewing sarcoma survivorship are lacking, particularly those utilizing machine learning.
• The Surveillance, Epidemiology, and End Results (SEER) program provides extensive cancer registry data.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting 5-year survival in Ewing sarcoma patients.
• To identify the most effective machine learning model for survivorship prediction in this cohort.
• To create an accessible tool for clinicians to aid in patient management.

Main Methods:

• Utilized data from 2332 patients diagnosed with Ewing sarcoma between 1975 and 2016 from the SEER program.
• Randomly assigned data into training (20%) and testing (80%) sets.
• Developed and compared boosted decision tree, support vector machine, nonparametric random forest, and neural network models to predict 5-year survival.

Main Results:

• The overall 5-year survival rate for the cohort was 60.72%.
• The random forest method demonstrated superior performance in predicting both cancer-specific survival (77% sensitivity, 91% specificity) and overall survival (83% sensitivity, 94% specificity) compared to other models.
• A web-based application incorporating the random forest model was developed.

Conclusions:

• This study presents the first machine learning-based predictive model for Ewing sarcoma survival.
• The random forest model offers a promising tool for predicting patient outcomes.
• The developed web application provides an accessible resource for orthopedic surgeons and clinicians treating Ewing sarcoma.