Prognostic determinants and predictive modeling in higher-grade glioma patients receiving radiotherapy: a retrospective SEER-based study.

Higher-grade glioma has an extremely poor prognosis, and radiotherapy is an important component of its comprehensive treatment. However, survival outcomes following radiotherapy vary across demographic and socioeconomic subgroups, and these disparities remain incompletely characterized. This study seeks to evaluate the influence of factors including age, sex, surgical approach, income, and racial background on survival in radiotherapy-treated higher-grade glioma patients, and to develop a machine learning-based prognostic prediction model. The clinical data of higher-grade glioma patients who received radiotherapy from 2000 to 2019 in the SEER database were retrospectively analyzed. To assess the predictive value of multiple clinical factors for 6-, 12-, and 18-month disease-specific survival (DSS), we employed Kaplan–Meier survival analysis, Cox proportional hazards regression, and six machine learning algorithms. Feature importance within the models was interpreted using SHAP analysis. A total of 35,765 patients were included. Kaplan-Meier analysis showed age, chemotherapy, surgical resection, income, and pathological type significantly affected DSS. Multivariate Cox regression analysis identified that an age ≥ 71 years and the absence of chemotherapy were independent risk factors for poor prognosis, while high income and temporal lobe tumors were associated with better prognosis. Among the machine learning models, LightGBM performed the best in predicting patients’ DSS. SHAP analysis showed that the core features of model prediction would change over time. These findings provide exploratory evidence for risk stratification and prognostic assessment of higher-grade glioma patients receiving radiotherapy.