Personalized survival prediction in pediatric glioblastoma using a machine learning-powered web tool.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
BACKGROUND: Pediatric glioblastoma (p-GBM) is a rare but highly aggressive malignancy with limited treatment options and persistently poor outcomes. Conventional survival estimates at diagnosis inadequately capture the evolving prognosis of long-term survivors. Conditional survival (CS) analysis provides a dynamic perspective but has rarely been applied to p-GBM. This study aimed to investigate incidence trends, evaluate dynamic CS patterns, and develop a CS-based nomogram for individualized prognostic prediction in p-GBM. METHODS: Using the Surveillance, Epidemiology, and End Results (SEER) database [2000-2021], we identified 597 patients (≤18 years) with newly diagnosed GBM. Incidence trends were evaluated by Joinpoint regression. Overall survival (OS) and CS were assessed with Kaplan-Meier and Cox models. Prognostic determinants were identified using multivariable analysis and least absolute shrinkage and selection operator (LASSO) regression. A CS-based nomogram was developed, internally validated, and deployed as an interactive web tool for individualized prognostic estimation. RESULTS: Incidence of p-GBM remained stable over two decades [annual percent change (APC) =1.52, P>0.05]. CS probabilities improved markedly with increasing survival time: the 5-year CS rose from 14% at diagnosis to 88% at four years post-diagnosis. Multivariable analysis identified infratentorial location, large tumor size, and disease extension as adverse prognostic factors, whereas gross total resection and chemotherapy conferred significant survival benefits. The CS-based nomogram, incorporating tumor site, extent, surgery, and chemotherapy identified through the LASSO analysis, exhibited robust calibration, moderate discrimination, and notable clinical utility. Moreover, risk stratification clearly differentiated high- from low-risk patients (P<0.001), while the web-based calculator enabled real-time, individualized CS estimation. CONCLUSIONS: This population-level study highlights the dynamic nature of prognosis in p-GBM, where survival probability improves substantially with time survived. The CS-based nomogram and online tool offers a novel, clinically actionable framework for personalized risk assessment and survivorship planning. Integration of dynamic modeling with molecularly informed approaches may further refine prognosis and guide therapeutic strategies in this devastating pediatric cancer.