An Online Calculator for the Prediction of 5-year Survival in Patients with Non-Metastatic Chondrosarcoma Using Machine Learning

Abstract

Background: Most prior studieshave used conventional statistical techniquessuch asthe nomogram to examine patient prognosis and survival based on the distinct subtypes and primary locations of chondrosarcoma. Nevertheless, few studies have examined overall survival in patients with non-metastatic chondrosarcomas. We aim to develop a machine learning algorithm to evaluate 5-year survival in this cohort and transform the most accurate prediction model into an online calculator for clinical use. Methods: Between 1975 and 2018, we gathered data on patients with non-metastatic chondrosarcoma from the Surveillance, Epidemiological, and End Result databases. From these data, nine features were used to develop four machine learning models, including the boosted decision tree, support vector machine, bayes point machine, and neural network models, which were then evaluated and compared based on discrimination, calibration, and overall performance. Results: Ultimately, 1202 patients met our inclusion criteria. With a c statistic of 0.88, a calibration slope of 0.997, a calibration intercept of -0.02, and a Brier score of 0.12, the Bayes point machine performed best overall, and it was integrated into a free, publicly available software interface that can be found at https://bayesglm.shinyapps.io/non-metastatic-chondrosarcoma/. Conclusion: Although this Bayes point machine-based prediction model has not been externally validated, the online calculator can be used as a reference tool for medical staff as well as patients on survival prediction. External validation of this prediction model should be the focus of future research to improve its credibility.