Abstract
BACKGROUND: Traditional survival analysis is frequently used to assess the prognosis of ependymomas (EPNs); however, it may not provide additional survival insights for patients who have survived for several years. Thus, the conditional survival (CS) pattern of this disease is yet to be further investigated. This study aimed to evaluate the improvement of survival over time using CS analysis and develop a CS-based nomogram model for real-time dynamic survival estimation for EPN patients. METHODS: Data on patients with EPN were collected from the Surveillance, Epidemiology, and End Results (SEER) database. In order to construct and validate the model effectively, the selected patients were randomly divided at 7:3 ratio. CS is defined as the probability of surviving for a specified time period (y years) after initial diagnosis, given that the patient has survived x years. The CS pattern of EPN patients were explored. Then, the least absolute shrinkage and selection operator (LASSO) regression method with tenfold cross-validation was employed to identify prognostic predictors. Multivariate Cox regression was employed to develop a CS-based nomogram model, and we used this model to quantify EPN patient risk. Finally, the performance of the prediction model was also evaluated and verified. RESULTS: In total, 1829 patients diagnosed with EPN were included in the study, with 1280 and 549 patients in the training and validation cohorts, respectively. The CS analysis demonstrated that patients' OS saw gradual improvements over time. With each additional year of survival post-diagnosis, the 10-year survival rate of EPN patients saw an increase, updating from 74% initially to 79%, 82%, 85%, 87%, 89%, 91%, 93%, 96%, and 98% (after surviving for 1-9 years, respectively). The LASSO regression model, which implements tenfold cross-validation, identified 7 significant predictors (age, tumor grade, tumor site, tumor extension, tumor size, surgery and radiotherapy) to develop a CS-based nomogram model. And further risk stratification was conducted based on nomogram model for these patients. Furthermore, this survival prediction model was successfully validated. CONCLUSION: This study described the CS pattern of EPN patients and highlighted the gradual improvement of survival observed over time for long-term survivors. We also developed the first novel CS-nomogram model that enabled individualized and real-time prognosis prediction. But patients must be counselled that individual circumstances may not always accurately reflect the findings of the nomogram.