Introduction: The Surveillance, Epidemiology, and End Results (SEER) database has been queried to include all malignant osseous spinal tumors, including chondrosarcoma. Machine learning (ML) techniques to predict disease have shown higher diagnostic accuracy than classical methods. We sought to develop and validate deep survival neural network ML algorithms to predict survival following diagnosis of chondrosarcoma, using a SEER database.
Methods: With the SEER 18 registries, Risk Estimate Distance Survival Neural Network (RED_SNN) was applied for modeling. Our prediction model was evaluated at each time window with receiver operating characteristic curves and areas under the curve (AUCs), as was the concordance index (c-index).
Results: Subjects (n=1088) were separated into training (80%, n=870) and test sets (20%, n=218). The training data were randomly sorted into training and validation sets using 5-fold cross validation. The median c-index of the five validation sets was 0.84 (95% confidence interval 0.7919 to 0.8677) The median AUC of the five validation subsets was 0.838. This model was evaluated with the previously separated test set. The c-index was 0.818 and the mean AUC of the 30 different time windows was 0.85 (standard deviation 0.016). According to estimated survival probability (by 62 months), we divided the test group into five subgroups. The survival curves of the subgroups showed statistically significant separation (log-rank test: P < .001).
Conclusions: This study is the first to analyze population-level data using artificial neural network ML algorithms for the role and outcomes of surgical resection and radiation therapy in spinal and pelvic chondrosarcoma.
Patient Care: RED_SNN may offer the advantage of increased sensitivity for certain adverse events such as OS. The use of ML methods may also provide a more flexibility and expandability to understand cancer progression in clinical practice.
Learning Objectives: RED_SNN is a valid method of predicting survival for spinal chondrosarcoma