Abstract
Brainstem ependymomas (EPNs) are rare and aggressive central nervous system tumors with unique anatomical challenges and poor prognosis. Traditional survival estimates offer limited clinical guidance due to their static nature. This study aimed to investigate the dynamic survival patterns of brainstem EPNs using conditional survival (CS) analysis and to develop a web-based nomogram model for individualized, real-time prognostication. Patients diagnosed with primary brainstem EPNs between 2000 and 2021 were identified from the SEER database. CS analysis was performed to assess changes in survival probability over time. Annual hazard rates were calculated to identify high-risk periods. Prognostic variables were selected using best subset regression (BSR) and least absolute shrinkage and selection operator (LASSO) methods. A CS-based nomogram was constructed using multivariable Cox regression and validated through calibration plots, ROC curves, and decision curve analysis (DCA). A risk stratification system and an interactive web calculator were also developed. A total of 697 patients were included and randomly assigned to training (n = 487) and validation (n = 210) cohorts. CS analysis showed that as patients survive longer after diagnosis, their probability of surviving additional years increases steadily. And six variables (age, sex, race, histology, surgery, and radiotherapy) were identified via the LASSO model for nomogram construction. The CS-nomogram demonstrated good calibration and acceptable discrimination, with 1-, 3-, and 5-year AUCs of 0.626, 0.649, and 0.656 in the training cohort, and 0.688, 0.692, and 0.687 in the validation cohort, respectively. DCA confirmed the clinical utility of the model. A risk classification system effectively stratified patients into high- and low-risk groups with significantly different survival outcomes. A web-based calculator was created to facilitate real-world application. This study presents a novel CS-based nomogram that dynamically predicts survival in patients with brainstem EPNs. By capturing time-dependent survival probabilities and integrating key clinical factors, the model offers a practical tool to support individualized prognosis, patient counseling, and follow-up planning in clinical practice. While it offers individualized prognostic insights, its clinical use requires further external validation.